# Day2 English transcript (concatenated hourly windows)

NOTE: Each block is ~last 30 min of recording at that cycle. Overlaps and gaps exist.


===== BEGIN RUN 20260917-0351 =====
involved and I felt like a lot of imposter syndrome actually when I was building it because I'm not a
coder. I'm not a developer and I can use cursor and I can ask you to do things but half the time
it's asking me to commit to GitHub and repose and I'm like I don't know can you help me and it will
help me it will do it but I am in there and I feel like an imposter. I actually recently tried doing
something like reproducing that thing on the left side, that video on the left side inside
Grapher and I was like hey can you just I set up a package alerts tracker. Yeah so inside Grapher
I recommend you do this because I don't I'm not always good at like knowing what packages are
arriving because I have my packages separate on it's like a separate email account so I don't
know like what's arriving and then I maybe miss it and then I'm home but the but the UPS driver
doesn't ring the doorbell and then it's all my doorstep and it gets stolen. Yeah so yeah
so I wanted a package tracker that tells me like this is what's arriving today this is what's arriving
this week and so it will check my email like twice a day and for like what's out for delivery
and then it will scan and then without for delivery it starts checking more often. Oh cool. Yeah
and then it even logs into my Amazon account nice and once it's out for delivery it will send me
a screenshot of the number of stops away my Amazon driver is. So you're truly getting like down to
the minute notifications. Yeah but I want it flipping and then also it checks like for door dash
it'll tell me like when my door dashes on the way so I've I've made this package tracker and I've
kind of connected like all these different all kinds of like food packages and also like
meal packages and everything in between. Yeah and then I connected it to my split laughter display
which is the best for which is the thing that flips and it's in my room and so whenever a package comes
it's like blah blah blah blah your package is here. Yeah I just can't I can't miss it. So you never
have to look at your phone like everything just gets notified to you on for packages. For packages.
I still have my flip. But for but that was so easy to set up because for in in cursor I had to build a
whole like like web interface connect to APIs like upload upload like create a web it was just
many steps and a lot of troubleshooting and then inside GROFA I was just like wait can you just
connect to my vestibule? Yeah I was like sure here you go. Give the API key and I'm like okay so I
pull up the app which comes with the split split flap display grab the API key paste it in and then
like two minutes later my my display starts flapping and I was like and it didn't ask me to build anything
yeah. You bit anything upload anything and so just the frictionless nature I'm coming up with so many
more ideas now because of that because it's just so easy to build things yeah and I feel like I don't
have to code anymore like the coding is invisible like it's no longer vibe coding it's just vibing.
It's it's vibe creating but yeah maybe yeah yeah yeah I'm not good at coding. I'm not
sure. Please don't like yeah please don't feature me. You know one of the things that I'm finding
that is really resonant about the work that you've done is that it is really shortened the creative
process but it feels like you also like have dug pretty deep like in your phone project you're kind
of like working with microcontrollers and like doing a lot of work with hardware and so I'm kind of
curious like you know have you used GROKBOT to help with sort of learning how to do those things or
you kind of like using GROKBOT as an educational tool and then also just like where do you get some
of that inspiration from because there's a lot of people out there who are like I don't know where to
get started right I don't know what to work on right I'm just curious kind of how you how you
like think of those ideas right yeah I'm starting to use GROKBOT to connect especially with tools that
have API yeah it's been really helpful actually the next step of my newspaper bot is I'm going to
actually take this this this this unstepped further and I actually got a robot pen that like actually
lots and draws it so it can draw like a full-sized newspaper okay and that can connect with an API
so I'm working on connecting GROKBOT to like this robot pen and so that'll be my next video okay
I'm still figuring out that one out so okay okay so so that's been really fun and yeah I just feel
like it has just been able to I've been honest I've been really manic the last
like I haven't been sleeping as well because I've just been I have had all these ideas but
in terms of where I get the ideas actually this this came from solving problems that I had so
last year I had a challenge for myself where I would not look at screens on Saturdays yeah I didn't
do it every Saturday but I would do it at least like once or twice a month it was very hard
but basically the ideas from like from Friday night until Sunday morning no phones no laptops
like I would I would watch a little bit TV but no like no no tiny devices no devices that are
connected to internet that could send me notifications and I did it because I was so burnt out
on like my phone and phone addiction and I was feeling just really not good like physically not
good mentally not good I felt like my brain was a sponge that held the maximum amount of water
that it could possibly hold and there was just no room in my brain for anything else so I did this
thing where I would not look at my phone for like a day a month or two days a month it was shockingly
hard because there's so much on your phone now like everything that you need is now on your phone so
it's much harder now to not to forgo your phone that feels like you're like giving up and appendage
you're like you're just losing contact with friends to meet up I can't like order food right I can
walk to a grocery store and cook which is you know that's that was a good thing for me to do maybe
that's also a good yeah yeah yeah yeah and so and then I also can't just look up something like
sometimes I'm just curious I want to know like well what time does the hardware store close
I can't look that up you know what extra what time is it exercise class today I gave up so much
in not using my phone so I had all these like problems and so I think you could identify a problem
or a pain point that you have in your life and then just solve it yeah so I came up with these kind
of ridiculous solutions which was creating you know this phone that would answer questions for me
on a giant split flat display and I think that would print my newspaper but the
the pump this solved actually was I know I'm not supposed to look at my phone first thing in the morning I
know that that I noticed when I look at my phone and first thing in the morning I'm scrolling
with like one eye open yeah I like my both my eyes have even opened yeah yeah and I'm like
I'm like scrolling Instagram I know this is not good from scrolling TikTok this is not good for me
and then I wake up and I start my day and it kind of met it throws my whole day yeah and then I would
have days where I was like okay I'm not gonna look at my phone I'm gonna go out in the park first
thing and those days will always be much better but that would take so much discipline because in the
morning I'm like well what's new who needs me was there an urgent email overnight was there a
calendar event that was moved was there something I need to know and so I would just
could not resist checking my phone yeah in order to get that information and then but then I would
see all my social media apps and I would see all the other stuff I don't need so I was like okay what
if it could just print just the information I needed which was how this idea for this morning
is amazing uh yeah well I'm sort of taking away from this and for those of you who are watching and maybe
you want some ideas of what to do with Grockbot thanks so much of what you can do is like you know
install Grockbot give it a problem that you are facing whether it's wanting to curb your phone
addiction or maybe getting outside more often or you know anything like that right like helping
manage a crazy schedule helping track your packages I think there's so many of these sort of
like daily problems that people face and you know daily things that you know if you haven't used
an AI agent tool a little while you don't you don't even know it's possible that you could like ask
an agent to help with some of that work and it sounds like you really sort of like you know product
ties that in some ways by saying like hey I'm gonna like go and figure out all these little things
where I can go and get Grockbot to help me out and I think that's just super awesome I can share a
few other I don't know about time yeah I think we're gonna be in more minutes yeah I can want
to do a rapid fight I can share a few other of my thoughts I've been really helpful so
um one is like when my favorite t-shirt there's a shirt that I wanted that was out of stock in
the color that I wanted and I just have like a back-and-stock tracker for items that are out of stock
and so once a day it'll go to the website it'll check for like the size and the color that I want
see if it's in stock and then sure enough like one day it had the alert and it was like it's in stock
there's only one I think I'll just want to return to and I went and got it immediately you could
also set it up to buy automatically I was going on vacation so I was like don't go buy it on
my back I won't do it. You're a package tracker you need to try it. And then my package arrived
yeah yeah and then my package arrived yeah um I have a shows tracker so um I just put in like I
just tell it like you're my favorite shows and they'll tell me when something is renewed for season
two when like it's gonna come out that day because I'm sitting down at night and I'm like what am I
gonna watch yeah and then this thing will tell me like hey your favorite show premieres tonight or
it just got renewed and that's you know that's something that I look forward to um so I think you've
also done a lot of proactivity right so like things that you'd otherwise have to go seek out
information for you kind of not have Grockbaugh pushing that to you so you kind of just have one
place we're like it's kind of a feed right of all the stuff you care about all these like
things that are arriving on your doorstep maybe some like notifications that would be over there
that you can have centralized so I really love that idea of kind of like putting everything in one
place and then the thing that I like really like doing is um I like Grockbaugh can control your
eye message I don't know if you okay so this takes I think a little bit of settings to set up
you have to try like change the settings in order for this to happen but Grockbaugh controls your
eye message so then to be able to put if then logic on your eye message has been so cool so for example
I was ordering an item to a friend's house that they needed and so I instacarded it to their house
and so I needed to tell them like this is when your item is arriving but I don't want to I wanted
to go take a shower like I didn't want to go sit and update this my friend so I was like okay
Okay we have a perfect conversation for you we've been working on the Grockbaugh team that's really helpful for us to hear
and like kind of putting my product hat on is what we do you have like what would you love to see in Grockbaugh next
maybe where some areas where you might want you know things to be a bit better or some areas where
you know we could add new stuff i would love for it to be able to
..
…
…
…
…
…
how much money I'm spending on my various subscriptions.
But the problem is things are always getting logged out
on the virtual machine.
But sometimes I'm logged in on my local computer.
So it would be good.
It's an Apple lifecycle better.
Yeah, that would be one pain point that I would love.
And you're sounds like you're logging into a bunch of services.
So you're probably hitting that pretty frequently.
Yeah, yeah, awesome.
Well, thank you.
It's like really helpful for us to hear,
especially from a power user such as yourself.
And so maybe just to wrap us up, first of all,
thank you so much for being here.
This is incredible.
I just find it is incredible to see the creativity
of what you've built with Rockbot.
I'm excited to see all the future projects you build.
I'm not going to ask for too many spoilers,
but the plotter sounds amazing.
That's coming.
I might need to buy a plotter and go make myself
a full page newspaper.
So we'll chat afterwards.
Just to wrap things up, where can people find you?
Where can people find your box?
And any parting thoughts you leave for the folks watching?
OK, you can find me on social media.
It's Karen X Cheng, CHENG.
And then if you want the newspaper bot,
it's newspaper.carenx.com.
Wonderful.
Or you can just search in Google like Morningbot newspaper bot.
Yeah, and we should get that in the marketplace too.
I think I'd give you a little bit of a quick.
OK, OK.
You can go to the marketplace and in the personal category,
you can go to the marketplace.
Bot slash marketplace.
And there we go.
All right.
Thank you so much for being here.
We're going to cut to a short break.
And then we will come back to building our three day
business and keep the process going.
Thank you all for being here.
OK, thank you.
We're back on.
And so that was a conversation with
Rich and Karen and Sharon on the Russian.
Yes.
And I honestly, one of my favorite demos for Rockbot
is Karen's internet-connected printer.
So if you haven't seen that one on X yet, highly recommend
I really want to build something similar to myself.
And so in the meantime, during our break,
I was sharing with Lauren.
Lauren did a bunch of whiteboarding,
started to think through some ideas
and directions for the game.
Yeah.
So we take a look at my screen real quick.
So I have TL draw open.
And I was basically, this is more for me.
I could have prompted a bot to come up with this.
But I just wanted to visualize it in my head how it would work.
But this is kind of a rough, I guess, a rough set of moths.
This is cool.
Yeah, you have like, you know, like the game screen,
the name of the game is obviously well working title.
Working with not a lot of you.
And then you go like a match screen.
And then just part of the captain.
Yeah.
Yeah, so I was thinking like, so you would pick a bot as your captain.
Right.
And maybe that has some implication later.
Yeah, right.
But now I don't know if there is any implication.
But you choose your captain.
And so that's the bot that you get to pick.
Right.
And then random one and then two of the ones that are up to you.
To assemble your team and kind of put the team together.
Yeah.
Yeah.
And then you can see the stats.
Maybe so once you click on choose captain, then you get like,
so this is like your list of bots that you've already hired.
And then you can pick which one.
And then I guess there could be like another, like an input here
that says create a new one.
Yeah.
But then you can also go to this team page.
And then you can manage the bots you have.
And maybe for now, it's just a very simple list of bots you've
had before.
And they're kind of associated with your account, right?
So when you logged in, you're kind of have a team that's so
interesting to track who meant to the bot.
The original creator or joke.
And it's exciting to create it.
Means anything.
It's just because it's just a person who upload it.
That could be a fun like a threat to me, maybe as a bonus
feature.
Do you like to whiteboard in the browser inside cursor?
And what is that unlock for you?
Can you do anything special with those?
Yeah, so the really nice thing about drawing.
So yeah, I have TL draw open in cursors browser.
And my agent can just see it.
So you can see, I just told the agent like, hey, can you
look at my browser?
I drew some mocks.
And it took a screenshot.
And now it has context.
Oh, that's interesting.
It's very useful context to actually go up and build it.
So I have it actually because our original version of our
prototype is kind of I still love this, by the way.
I think this looks pretty great.
I mean, it doesn't look like a game.
It looks like some kind of trading cards.
Maybe they're trading cards.
But yes, I wanted to make it look more game like.
So that was the inspiration for how do I actually want this
to look?
Because this is not very, this is not the game I want to play
personally.
So you know, like very big and minimalist and fun.
Yeah, I like the big sort of like take over the screen, really
focus on the bots.
We can maybe bring some of the cards, the CSS that you were
calling out to bear here and have this really be simple.
So you see your team.
And then did you have a chance to prototype out how you'd actually
go?
And then, yeah, so once you pick your captain, then let's say
I picked Dr. Eggbot as my captain, then you get to decide
the order of the, and you use drag and drop.
Cool.
Yeah.
Then you confirm and then you put, you get matched, you get
a finding match screen.
It'd be ideal if the match making was so fast
that you've never saw the screen.
But maybe it takes more than like a few seconds
to show you the screen.
Cool loading animation, maybe.
And then, okay, so let's say I got matched with Maddie P,
who has Madbot and round one.
So unlike the prototype we saw before, which is very text, text
wordy, you just see like your one bot at a time.
And it's like round one.
And I'm still kind of thinking about how this would look.
But I mean, me, I'll get the bot to kind of now that I have
a one basic mocks, just get the agent to actually riff on these.
And so I was just thinking, okay, random to use Riz and
Matt bot used to implode.
And I was talking of this idea of like not making it to stat
focused, but that makes sense because it's like,
it's not very fun to me, like random to use Riz.
And that was 37 hours.
Like who cares?
So thinking like maybe there's like a way to visually show like,
well, this has two arrows of power, you know?
And so you can, it's more like conceptual.
And an implode all only has like, maybe this is just based on the
stat.
Yeah, I think that could actually, that could actually be fun.
It's like we just maybe we chunk the stats into war categories.
Like 0 and if 20, 20 to 40, 40, 80.
Right.
So we just show we draw numbers are like behind the scenes.
Yeah.
So the people who really want to dig into the numbers and like,
you know, get into it, they can.
But if you're in, I don't know, I feel like the numbers,
or maybe the numbers are visible.
I feel like the numbers could be hidden.
Yeah.
And then you will see how that feels.
That's a good point.
We can always add that later.
And maybe yeah, it could say, I feel like the number is only
interesting if you can set power.
Have an impact on the number.
Yeah.
So I think that we just show the relative strength.
Yeah, I like that.
We should just take, because I think we go up to what 150 max,
150 total stats today.
So it's probably way for us to be like four.
Yeah, you go four.
And it like, it'd be like a cool thing.
And maybe you don't even know that your opponent bot is like,
like, it's revealed.
It just gets shown when the user that's cool.
Extremely powerful.
I really like that too, because it sort of amps up the drama.
And I feel like that's something that keeps you really, you know,
keep people engaged.
I would love that.
I'm like waiting with anticipation to see what happened.
And each one of these, can each one of these charts and so.
This part definitely needs some design help, I think.
I'm not that good at coming up with designs here.
But just thinking through like, how do we do these?
I'm going to make the animations look good.
How do we make the, how do we just make it not like a tech suit game,
where it's just a tech sweating around everywhere?
So some cool animations.
Okay.
Like a win and lose animation.
But yeah, that's what I just sketched out really quickly.
I love that.
And we have our, oh, there you go.
We have some marks.
I guess I didn't ask it for marks, but maybe I just need to refresh.
Oh, there you go.
Cool.
Okay, let's try.
Let's walk you in.
Hold on.
Is this a real log in?
Okay, no, it's fake.
Thank goodness.
Okay.
There's a fun logo.
Okay, so I'm diamond.
Cool.
Yeah.
What is diamond?
Oh, so I had this idea.
So since we have a hero ranking.
Yeah.
I thought it would be cool to show instead of again, like instead of showing numbers.
It's far more, I think, engaging and interesting when the player doesn't see the actual.
You don't have to see the guts of how the game works.
It's kind of, but it's more, you know, instead of saying like, oh, my, my hero is one, three, seven, two.
Yeah.
I'm diamond.
That's like, we cooler.
Yeah, like I'm in this cool category.
You're like, you're like, brawn.
Yeah, yeah.
Yeah, yeah.
Yeah.
Yeah, right.
At this game.
So I like that.
Okay, this one doesn't really look like my mom.
Is it still working?
I like the big loss.
Oh, that's kind of, I like the summary screen here too that it came up with where we'd show all the boss and the summary screen if he's go down.
Oh, interesting.
I can show you something.
I guess maybe that's interesting because then you can like tweak your strategy.
Yeah, that is one benefit of actually doing, of showing the stats is that you could kind of then show why you lost.
So like in the actual battle, maybe you don't know the stats, but afterwards you're going to get the the game summary.
And that would actually be really cool to share on social for people too.
I think.
Yeah.
So I want to actually tell you that some feedback here.
Like I draw.
Maybe I'll just do the shift quick.
Oh, that's what you're doing now is using design mode and cursor right to kind of give direct feedback on the actual part of type.
There you go.
Yeah, you have shift click here.
This doesn't match my box.
I was giving the agents feedback on what to adjust three.
Big cards where the first card is the captain selector, I guess.
And the other two cards are.
Okay, and then that's actually it was just for the sake of the scene to fill.
I'm just going to go with the link she's bought.
And okay, so now we can.
Short the order.
How me out.
What do you, what do you think we should go first?
So for this one, I think let's do our, well, I can't see what.
He told gets charisma.
I think we should put our like critical first that feels like it's a guaranteed win.
Or yeah, then actually implement the track.
Okay.
So I think what does this.
Oh, that was not.
This thing is blocking my screen.
Oh, I guess.
I think you might want to be lose your screen.
Yeah, and we're back.
I think I'm going to give you back again and say like.
Keep pressing.
Make this dragable.
To the order.
And you can remove the order button.
And I think that might be cool too is if we want to push some of this up to.
I did push.
I did push it up so you can.
You can certainly do.
Awesome.
Yeah.
I was working on some like landing page stuff based off the design Lauren has.
So.
I'm just kicking off.
Okay, then I'm going to confirm.
Okay, finding match. Check it out.
Finding a match looking for another thing.
Okay.
So this is just like a demo.
Yeah, so it's.
Yeah, this is kind of cool.
Yeah, but it happened too fast.
Like I think.
We need to add the animations.
Yeah, but you know what?
Maybe this is fine. We'll ship it.
Yeah.
And then we can iterate.
Exactly.
So.
So link she's engineer bought used.
Riz.
I don't want to show the details here.
So I think we'll give it some feedback as well.

===== END RUN 20260917-0351 =====

===== BEGIN RUN 20260917-0515 =====
the code base. So when I'm live in a call instead of you know needing to get back to somebody,
I can just ask a question live in a call. Like one example is we work with customers
to set up cloud agents a lot so I could ask like hey can you send me a customer facing answer
on how to set up cloud agents. Also send me a few outcomes that amplitude and fair achieved.
So like maybe for this question I would have had to like talk to an engineer or bring an engineer
on the call to help walk the customer through setting up cloud agents and then maybe I would have
had to like talk to the amplitude and fair account managers and understand like how they're
leveraging cloud agents or talk to those customers myself but it'll pull that customer or that
response from a code base. A lot of times I paste this in Slack so it kind of already knows how I
work and like shares with me like a paste ready version I can share but it gives me all the steps
to set up cloud agents and then it also shows customer outcomes like how have we worked with other
customers to set up cloud agents and what are the outcomes achieved. So I use this a lot and then
just like pulling content. I think like the most powerful thing for me is just like actually
watching content and doing things. I like to think of Grapbot like not just like as a thinking
partner or like a research partner actually like a doing partner so I could have it you know watch
a customer webinar with a CEO and be like okay based on this webinar draft me an outreach
messaging or particularly the messaging from my CEO to reach out to the CEO instead of a meeting to
talk about their three key objectives and how it relates to what we can do for their business.
So when I first started working with Grapbot I think I mistakenly started using it as I would
like a chat interface so I was like hey go and do this research and then it actually gave me the
links to go and watch these webinars myself but I kind of push back I was like pay no go watch those
webinars for me and draft an email for me. So I think you have to change your mindset with how you
work with these like think of them like colleagues who have their own computers they can work overnight
and they can do these things for you so a lot of times when I'm prospecting I will just have my
prospecting bot work overnight on like a very long list and then just have all the emails ready to
go for me in the morning and then I think every day I like to add like you know 20% more I'm like okay
well you drafted these email responses that's great we'll go in draft like LinkedIn responses or
go find 10 other people to reach out to so you're gonna build upon your bots as you continue to use
them. I also use Grapbot a lot for travel and expense I'm sure we all do also expenses are not
fun so you can find like Starlink flights which is helpful and then I use Xbot a time so this
is connected to the X API as I mentioned like I found just like LinkedIn is a pretty out of date
with company information and you can kind of dig and find like who you can reach out to and then
I also just like learn it like use a Grapbot for learning so if you're like researching like a new topic
like one that like to like think about is like SpaceX or maybe it's like learning about
or like a topic like I want to research new restaurants or something but you can just like pull
up a new chat and research restaurants reach out to but yeah these are all my favorite things
you can ask it to be like hey can you watch the most recent webinar with Roman around Grapbot on
Lenny's podcast. I think we have a lot of Grapbots actually watching this as well. Yes yeah this
is how we Grapbots live so it'll go off and watch any live Grapbot webinars. Yeah and Chris I
want to hit on one thing on the engineer piece the engineer one this has been impactful for me as
well in terms of like speeding up my deals. How many times we all been there where it's like oh yeah
let me find the answer and get back to you but now with the engineer bot we've been able to like
quickly get those answers answer them on the call and really speed up the next steps in an opportunity
and that one I highly recommend and really impactful for us and organizations as well.
Perfect I think that's all I had planned for today happy to go through any questions happy to
go through any demo items but would love to you know chat more we'll kind of finish up with
what we've learned today and then we'll open up for questions at the end. Sure I want to switch back
there. Great so a few tips and tricks kind of what we learned today looking at that as well as
overall the biggest thing we talk about our customers is really connect your stack like look at
what you use in your day-to-day operations and get those connected directly in Grapbot that's
really going to help you speed up the time to value here and it's really impactful in terms of
you know getting those aha moments early and starting to really you know work within Grapbot.
One thing we've talked about too whether you use like a cheap of staff to delegate down or
Grapbot by Grapbot you know have one bot with one job like this is your focus and Chris showed a great
example like we have engineer we have pipeline generation travel expense have them be an expert
in that field these are your team members that you're onboarding have them be an expert and give
them all the relevant information materials to be great at that at that job and then another one
is work on the routine some people use them more regularly and some people don't I frankly like
the automations and routines can I grant a pay like set it and forget it have it do the work
and you know come back to me with anything necessary and that's a good way to make sure that
it's always working and doing the the tasks needed so three tips to take you forward on your Grapbot
journey after this but would be great to open up the four to any questions you may have for Chris
and myself sure I think it depends how you're using it like there's ways to reduce your amount
of usage which is like running less routines a day I think we had one example where someone acts
like a routine for every 15 minutes so those are like ways to reduce your usage I personally don't
use routines a lot because I find them too noisy so I just run them once or twice a day and I think
another way to save on token consumption is I personally like work with my specialized bots
and I feel like if I use my chief of staff we I didn't show this today but your chief of staff can
orchestrate other bots so I've been working on like a deck it'll call like the you know deck designer
or maybe working with a customer it'll like call that like customer bot and like pull the account
data I personally like to work with like my specialized bots and I found that reduces tokens as well
but it's highly personalized to like what use cases you're leveraging another tip is I use computer
use if like there's not like a good MCP available I think like we most of us work in sales and I've
just found like a lot of the sales tools we have don't have great MCPs but I would save the computer
use for if there isn't an MCP available already thankfully we shipped a lot of MCPs out of the box
like I think you know we shipped a gong granola sales force and more so use the MCPs when
you can and save computer use for when you need it great point just to add like I'm probably
the opposite where I'm like chief of staff routines so perhaps my my my spend is it's certainly higher
but one thing I make sure to do is like ask rock bot for optimization tips right it's a great way
to go there and say like look like figure out a way to be able to optimize some of our workflows here
and what's the best way to you know save money the fact that we leverage first party models and
a lot of these things really helps reduce that cost anyway thanks you hear me yeah what do you think
it's a good hack in order to maintain a really good knowledge space like connected to notion granola
asked to download of CDN and a management of CDN account or I mean what it will be a good hack
if you get beer and beer knowledge space yeah I get that yeah start that I think I think
finest are correct like the we meet where you are right like what are you using internally
within your company and where does that knowledge sit that's typically the best way to be able to
quickly get access to start to work through these things quickly for us I feel like everything lives
in notion and slack and so connecting those two has been really the unlock for our company in order
to like hack if you will way to move quickly yeah obviously we use sales force so sales force has been
like a like a source of truth but I think I see rock bot as like an orchestration layer of all
those tools so whenever I need data like I can pull it from rock bot so whether it needs to go
to data bricks or sales force or notion or slack I don't think anyone any company is good at
maintaining data and having it in one place usually at least before I had rock by it was like oh great
you want this data like go to data bricks oh you want this data go to sales force or this is a notion
or no someone just posted from slack the other day so it's so it's like really impossible I think
especially for a very slack heavy companies to stay on top of where data lives or when something
changed so I think rock by is like a really great outer layer of pulling that information wherever
it lives whether it's notion or sales force slack or anything delegate tasks to cursor or
get the tell it to hey this is what I want and then go build with cursor yeah you can tell rock
bot to spin off cloud agents so you would just have your cursor account connected but if you're
like hey go spin off cloud agents and cursor you would just tell the bot to do it in natural
language right thank you we got one question from x there's some confusion on when to use a
specialist bot so the question is would you set up a bot for a major client or would you
will create an expert that handles all of your client I think this is preference I think I even
asked my growth bot the other day I was about to spin up a new bot and I was like hey like should
I keep these bots separate or should I move it all to one space so I personally have bots
for each my strategic customers but I only work across a couple where I know other AES
that other companies are even at SpaceX AI work across hundreds of customers so I think it's
preference but what would you recommend yeah absolutely preference there and the way I do it is
somewhat similar but different where I've got bots that are delegated to these customers which
is looking at their intent you know kind of taking the notes from our previous calls and able
to kind of feed into there and then I've again got my chief staff that kind of directs them
on what to do and then I've got the intent one that's actually just looking at all things
rock bot and building demos for me in my behalf on the rock bot so that any call I get into I can
quickly spin up a demo and go from there one of those things where it's like once you get started
with rock bot you can kind of figure out the best way to do so and I think x is the best way to find
those quick wins as well there seem to be some pre-packaged skills already right so if we
download these can you customize them for your use case and get it those as well yeah one we
already have some sales use cases on our rock bot like template marketplace but you can also
create your own so if you want to share your own templates with your team or even on x like I think
everyone in this room probably has a good template you can steal or share but there's some on the
back of the cards if you're in the room with us and we also have a marketplace online I also have
an x account so I like to share mine but yeah you can steal others templates and I think that's
what makes rock bot like so fun and easy to get started is you can steal what somebody has already
built and customize it for your own use cases team and the product team from the sales organization
is it via slack or there is a better way to do that from rock bot like how do I how do I communicate
with other departments talking to other organizations sales is talking to marketing all the time
product all the time or the like you know so what is the best right now we do it of course
through slack but is there a better way to do that communication yeah right now I do that via
slack but you can also use rock bot to track channels so if you're in a marketing channel maybe
about this event and you want to track what's going on you can keep track of that or that event
like within rock bot so we've been talking a lot about personalized setups your preferences versus
yours how would we set this up for or do you work on setting it up for a team where you kind of
had somebody else in sales here's your kind of standardized setup and then personalized from
though are like leveraging a power user that's done really well like Chris in this instance that has
templates and then kind of sharing those out for as a sales rep at your company I'm going to get
these five bots populated into my instance and I'm able to run from there and if I'm in marketing
perhaps like another five so possibly it's really endless there and that's been a big big
shift we've seen more recently of like having this pre-enabled based on your your roller personas
in the company so is that set up through your cursor account where you start setting that up on the enterprise
level or yeah directly through rock bot you can kind of you know build these up and then share them to
new users right so a little bit like let's build them out and then a new user will just share them over so
they have them day one okay are you able to demo the echo bot on how the doing call right somehow you have
some streaming capability to stream the transcribe to the bot itself so that you can understand context
to update the slacks or something this is how the echo bot say you have
hi sorry how do you say how do I stream the echo bot oh echo bot oh okay yeah so I use granola
so I'll do discovery usually like most of the call and I just like stop the echo bot or stop granola
and I run echo and it'll just pull the granola or transcript automatically and just like
start working on the slide so granola is like very quick um I also use gong too so if you use gong
I would say use gong because it takes a it takes a couple minutes for it to load after the call finishes
so that's probably more for like the next call or to send out after the call but you can run granola
during the call stop it and then just run the echo bot and it'll live update during the call
I think it takes a two minutes give yourself two minutes and then it'll update
got it thank you yeah perfect um I think we're going to wrap up here but lastly uh thanks for all the
questions uh before we jump into uh building I wanted to announce that it's a challenge on
x so you can share a grok bot we talked about temple it's a lot you can share one of your grok
bots that you can't live without and how you use grok bot for work um and you and a plus one could
head to starbase uh and actually get the chance to watch a starship launch in person
jambary jealous I would love to watch a starship launch so to enter you can follow at grok and at
bot on x and then quote our challenge post on x with the description of your grok bot and a link
to your shared bot template and uh you can scan the QR card card here and save the x post uh you'll
need to quote to enter the challenge uh so best of luck great perfect thank you all
we're back and while loran and roshan are off getting some work done i am joined by
another mat one of my favorite mats
that can maybe tell us a little bit about himself and we're going to go into how we use
grok bot some of what he's seeing in ai but first i'll let him introduce himself
sure everybody i'm at berman i make content for a living now which is still wild to say out loud
yet all about artificial intelligence i do everything from model reviews breakdown news sometimes
geopolitics research papers sometimes tutorials but uh everything is uh generally artificial intelligence
and uh been doing it for about three years before that i've been in technology for about 20
nice amazing and while you're number one a friend of the grok bot team we've hosted pop-ups
with you we've done a lot of collaborations also just a really great content creator and i
thank you take some inspiration from your work we've been friends for a while i remember you're
talking about moving up to the bay area and now you're in the bay i see you around all the time it's
great um how or i guess we can start with how are some of the ways you're creating content changing
with ai and then we can tie that into it to some other thing but i think

===== END RUN 20260917-0515 =====

===== BEGIN RUN 20260917-0554 =====
cursor and then eventually SpaceX like right around the time of this acquisition and so there's like a lot of context going on
And luckily grok ball was like in development at the time. So I had access to it throughout my entire tenure here
I just use it for onboarding like Lee hired me shout out Lee amazing amazing guy Lee Robinson
Yeah, at Lee Robb on Twitter this little plug and you know
I feel bad like being somewhere and I have to like ask people things like I feel useless
But even then normally at a company you start your job
You're like, okay, like I guess I got to open the company wiki and like read all this stuff
And then I don't know where the stuff is I got to ask somebody I got a putz around and search look in slack find messages
It's like the information retrieval very similar to how you're retrieving information across your email and your calendar with your wife
All this stuff is like now you actually have the shared context that you could just say
Who owns this thing look in slack look in notion. How do I get XYZ done?
Oh, actually I'm blocked on perms like have other people been blocked on similar permissions in the past
Who did they talk to and to get this stuff done totally and you just like what I've found is the number of슷s
I've had to send to like a manager or like to my teammates to ask them random questions is like nearly zero
And that compounds over an entire organization if you have everybody that has the agency to be able to just answer questions
Okay, Matt, let me extend on that point because you're right
You kind of made me realize how how valuable that is in the moment with AI like if okay if I have
You know 150 emails coming in every day and I'm trying to remember in a moment. What did this person say about?
You know a sponsor. What are they? What was the last context? I had with them and I'm working with a bunch of sponsors
I don't necessarily have to go search through my email dig through it
Look through the thread look through all of the replies. I simply bought and it just it'll go and find it
Yeah, and so I think that
Reduces the amount of context switching but also context searching which is what you just mentioned and
That also is a huge huge value like just the and okay now multiply that across multiple email addresses
Notion and slack and you know, yeah drive and all these other things now you have
documents and context in so many different places and
You don't have to remember where each one is now, right? And I'm
Realizing like I've been doing that, but maybe I didn't put it
Verbally how how important and how valuable that that kind of piece of it is yeah
And I mean to me these are like universal problems and it's almost like I can't really remember what it was like when I was like
Oh, man, I need to like go like look through slack
I'm like I'm never gonna do that again because it's like really hard to find stuff
And searching across multiple calendars. I remember, you know, just a couple weeks ago
I got invited to something I think it was like a company event and I was like wait
Did someone post that that was happening in slack or post that or send me an email or just put it on my calendar
And if they did put it on my calendar, like when was it and you talk about like hey
We just want AI to be easy right like as humans
When you're overloaded with all this information and to just ask AI
Could you remind me like when this thing isn't where it was might seem like a simple thing
But that could be the difference between you missing like an important meeting or being there, right? Totally
Do you remember Microsoft Recall?
Do you remember like maybe a year and a half two years ago where Microsoft was they announced this product Recall
Which would watch everything you do on the computer and it was kind of this like super powered search
So and so it would effectively like take screenshots of everything you were doing on the computer
And then you would be able to ask AI like hey, I remember this email like we're with
They like leading up to the launch got so much flack
About like the privacy and the security and and it's just it's like
It's it's like very different from from how grokba works
But I think back to that and I at the time was so excited
Yeah, just not having to remember where all the things are yeah in in the 70
You know the 17 different places that I stored documents and and so
Now he's kind of opting in I'm like no, I want to share all this. Yeah, I would have been okay
But yeah, the time they got a lot of crap. It is interesting
It is interesting as like leaned into kind of like some of these agentic
You just just gonna work out. You trust the guardrails. Um, also security has evolved like substantial and yeah
It's not taking it's not anymore. It's like no
We actually have a secure connection to the system
I mean I mean even like some of these purpose for mentioning earlier
This year I remember you know because I do technology demos
Yeah, someone was like hey
I want you to like demo this thing on our service and I look at it
I'm like I can't show people this like it's just dumping passwords in plain text now
Come here. That's a grokba where it's like you have secure entry
It goes into a vault like the agent can't even see the password, right?
So that is much more much more secure and you have a much smaller blast radius
If any of the systems get compromised
Shifting gears a little bit one other use case you taught told me about which I think is interesting
You told me grokba helped you with your electricity bill. Is that correct?
Yeah, I'd love to hear about this. Yeah, I pay a lot for electricity
So, okay
So I kept getting these emails from PG&E saying like hey you could save on your electricity bill
You have an electric car. So you should you know think about one of these other plans
I looked at the email first of all it's like three pages
Okay, like a ton of text and and maybe maybe I have no attention span. Maybe that's it
But I'm looking at this email and I'm like, okay
I would like to save money and I would like to change plant and then I would just like
Oh, then I have to compare like what a plan should I be on they don't tell you
And so I had so I gave grokba
I took a screenshot of the email gave it to grokba and I said figure out if I should switch plans on which one
Look at my previous 12 months of electric
electricity usage
Look at the bills and and tell me and so all it asked me to do was log into PG&E
And that was it and so it did all this review looked at all the previous 12 months of my bills
And said like, oh, yeah, you could save a thousand dollars by switching to this other plant that is
You know, especially tailored for having an electric car and doing like yeah off peak usage
charging off peak usage power
Um, and I was like wow that was like really easy. I had it double check the math twice just in case. Yeah
Yeah, I switched plans this morning. There you go. Yeah thousand bucks thousand bucks a year. I mean that
It's incredible. Maybe
All in all took about 60 seconds of my actual mental attention
And right I love classifying problems like these right because it's like
Again, you're talking about like a very specific personal problem and there are my p people out there
They're like, oh, I don't actually pay that much for electricity like this doesn't apply to me, right?
But it's like there are all these things in our life that we have to pay attention to
And part of being an adult is that like you can't it's just like so hard to give your attention to all these things
Whether it's your like your kids emails or like stuff at your job
It's like you're so maybe you're so overloaded that you pick a more expensive supplier at your work
Just because it's easier for you to just don't want to think about it. Exactly. Yeah
And now it's like, okay
Well, if I can at least map the problem space and then put an agent on figuring that thing out
You don't have to worry about that right and you can optimize these things. Yeah, it's so interesting
You say that I've talked about this a lot of my channel
the ROI on
tasks
That had kind of a lower value are is now
Substantially greater the ROI on let's say like
You have two things to choose from one which would have been $10,000 of value and one which would have been $1,000 of value
Obviously, you're going to focus your human mental attention on the $10,000 one maybe
Forgetting or just dismissing the $1,000 one because it just didn't make sense. Now you can do both
Right and both of them take substantially less attention and and thus like the ROI on investing in that
getting, you know, whatever the $1,000 value thing like
Hey
Like simply getting changing plans on on my electricity or whatever else it is
It makes a lot of sense now
It just becomes much easier and I think an extension of it is just have it go through your email and look at all your subscriptions
Like even if it's $9.99 a month and you haven't used it in six months go ahead and cancel it
Yeah, and you can have grok by you can have your ai assistant do that for you now
Yeah, and like previously you would have maybe said like okay. That's $9.99 a month. Maybe I'll use it next month
I don't really want to click through and have to answer 17 questions about why I'm canceling
It's like no no grok by go ahead cancel. Yeah, and that's it. It's done. Yeah, totally great
I mean and it just incentivizes like the agency like if you if you're the type of person that you like to get stuff done
Or you just feel like you don't have enough time, right?
You're limited by time or resources to get things done
But you have like the energy and willingness to like dig into problems. Yeah, it's like, hey look
I mean anytime you've ever gone to the doctor's office. This has happened to me
I use grok bot and and grok for these sorts of things like you get a bill and the bill makes no sense
And you're like dude, where did this number come from? Nobody can explain it to you
You can just use AI and not have to spend your Saturday morning trying to understand a bill
And that's like you said one of those things where it's like, oh, maybe I just don't want to deal with this
So I'll give him a couple hundred bucks and it's just like
You know totally now I can have something that that helps me like figure that thing out. Yeah
Yeah, I want to bring up something. Okay, maybe you struggle with this now
I I am getting more work done than ever and I know you've probably heard this
I'm also busier than ever. I'm actually struggling on the attention
Or like not the attention the context switching between all of these things
I'm kicking off so many different agents and they're coming back
And I still like there is still this cognitive load of switching between different threads switching between different pieces of context
How do you solve for that? Like what what does your workflow look like? Oh, that's interesting
So it's funny. We actually talked about this in a session, uh, like on the stream just like earlier today
Um, and Lauren, uh, who's probably sitting over there somewhere hacking on our uh, our software game design
So you know shout out potato potato gets all the shout outs on twitter like you cannot if you go on twitter
Just it's just potatoes. Yeah, I've been seeing her name everywhere. It's just people saying potatoes and say, um
So one workflow she mentioned which is like she uses a like chief of staff
She just calls her steve which I like better than chief of staff chief of staff is like overplayed in my book
Like we're on the stream if the chat in the stream no more chief of staff
Can I show you I have chief of staff? Of course you have chief of staff
I'm switching it right after this now. I have been shamed into switching it
I just feel like, you know, we we have to we have to be more creative
We have to be more human with our with our other agents. That's so anyway
Just basically like the same model as a coding agent
Which is that you have like an orchestrator and then you can like fan that out into subagents and rather than interacting with the subagents
Um, you interact with the orchestrator. So you could be like chief of staff bot talk, you know
So, okay
So this doesn't work without one other piece which I'll describe
Which is lauren has like a bot factory bot, which is she
Thinks about how she likes to create bots and then she put that into a bot
So chief of staff bot talk to she calls it doctor egg bot. This is all very
Steve talked to doctor egg bot right and create a, bot to
I like that I I've done some of that but I find myself going into each individual thread. I'm a huge fan of having things separated by topic
And so I I do find myself like the chief of staff is where I go in the morning
That's where I go to like really think about things and kick kick off the day's task
But as each of the individual bots are working, I do find myself going into the bot
But as each of the individual bots are working, I do find myself going into each one
And and it's kind of interesting. I do think eventually I'll kind of more and more try to push things to just being summarized really concisely in the
I'm just going to call it chief of staff for now. I'm sorry
But like I'll probably try to push more and more things
My attention specifically into the chief of staff rather than trying to go into each and every single
That I have I mean, you know the other thing that I'll say is because I feel like everybody in AI feels like they have to have some sort of like
Swab answer to questions like these
It's just like overload sometimes like as an engineer if I'm writing code I have 10 cloud agents running at the same time working on 10 things like
Writing code used to be like, oh, I'm going to like put on low five beats or whatever and just like chill out and think about one thing
And now you're you're just like an overloaded manager at all times. So I think like as an industry
We just need to evolve to understand the way to handle some of these problems
And I think people I've talked to like everybody's like a little stressed out and trying to do more, you know, and that's normal
It's so nice to hear that
No, it's nice. It's nice because I I'm not I'm not alone on this
It's just true like mine myself my entire team were getting more done than ever
We're busier than ever. Yeah, but there is this like cognitive overheating at times where I'm just like
I need to step away
And I think a lot of these are probably design problems that'll get polished over time
But um, it's really nice to hear
Yeah, no, I mean suffer with it too and and I think it actually uh in some ways like being busy all the time
It makes you think like um
Like what are the actually the right things to focus on right? So it's like, okay
Do I need to do this all the time? So, uh, yeah, I think uh, this was a great conversation. We have a customer
Video that we're going to kick to in a second here Nokia and they're analyzing 50 million lines of code in two weeks
With cursor. So we're going to kick it over to one of our customers. We'll learn a little bit more about that
cool, dude
That that flew by I can just keep going with you
Oh, whoops
Oh, we're back
The SVP of product and engineering for one of our product development divisions at Nokia
We are trying to leverage cursor as an agentic platform to introduce
specialized agents that are persona specific
For every aspect of work that we do that can partner with the human in-seat that does that work today
Those things become much more possible if we are able to iterate in much much faster loops
That directly impacts customer economics
Uh, lower's unit cost improves customer satisfaction
And then finally there's an entire class of things that I think we're now going to be able to do faster
That are to do with increasing resilience reliability and security in the network cursor and uh, curses agents
That are increasingly many of them are out of the box is
a massive step up in
Accelerating analysis getting insights that otherwise would not be possible. So we're at the beginning of a very critical journey
We've burnt the boats our plans our goals for this year not next year for this year
Hinge on this working. We have a product line easily 50 plus million lines of code
We are I would say in an intermediate stage of
Decomposing what is fundamentally a monolith oriented architecture
It would be a very daunting task for us to be a very daunting task for us to be a very daunting task for us to attempt to attempt a manual
Inspection of that code base. We have undertaking that with cursor our early results of that analysis have been
Remarkable we have been able to do precisely what I just described
Which would have required the spoke tooling
Would have taken far longer probably several months of effort across probably a dozen or more expert people
We performed the initial analysis as one example within a matter of I think a couple of weeks by
Two people we've seen evidence of cursor being deployed very successfully
In situations that involve a need to move very quickly where we have run into a customer impacting defect
That might be having a subscriber impact and now we're able to do that against an extremely large corpus of service requests
and radically accelerate
something like a root cause analysis these
Examples have not been one or two. They've been quite a few and growing in number
That's where we're seeing I'd say the early success and where we would love to take it next
is leveraging
cursor as a platform for
multi-agent
Deployment and orchestration where really the role of the engineer the architect becomes
The supervision of agents that are calling out to tools that are performing work at scale collaborating with one another
We are
literally
redesigning our entire software development life cycle from inception
Through product delivery and everything that comes in between
And we are back. So matt has left the studio now only one rep matt remains and we have another we have another guest
Gotta gotta keep the great guests coming. Um, yeah, welcome back for those of you who are just tuning in
We are in the middle of a three-day live stream here at grok bot galaxy in san francisco
part of dream force right next door to all the festivities happening right outside
We are building
Together over the course of three days a business entirely implemented by grok bot and coding agents
And so we've spent our first day kind of working through some ideas today. We're really deep in coding
We were earlier today showing you how we're using bots to drive a lot of the core workflows that we use
To power our software engineering stack
Lauren is currently off stream working on reimplementing some of our prototypes into a
Scalable game system and so we're really excited to show you that when we come back on in just a few minutes here
With the full team in the meantime. We're really excited to bring on our first intern
Yes, we actually just hired an intern
I mean, I feel like your team needs an intern right like got someone who can help out
Yeah, um, but but you're more than just an intern. You're also currently a student at stanford, right
You wanted to introduce yourself and say a little bit about who you are. Yeah, so everyone
I am the first forward deployed intern
So on my full-time job, I am bringing them coffee on a day-to-day basis
Currently actually before this live stream started. I was handling all the coffee logistics really really great
But in my free time, I guess whenever I get that
Um, I'm a full-time student at stanford university. I'm a rising junior and
I need a job
Hopefully we can help with that and hopefully this job is some good on the ground experience
And what are you majoring in at stanford? I'm majoring in and right now
Cool, and you are a huge grok bot user. Is that right? Yes
I personally get through all of my usage and half the time limit. So okay, we should really maybe take that feedback back to our usage team
To figure out how we can make that a little bit better or just give me more use
That would also work. Maybe there's an intern benefit that we can see. Yeah, some intern credits. Okay. Okay. Very cool
Very cool. Well, thanks so much for hanging out with us
You know, I think one of the things that we've been doing throughout the stream here is
You know, in addition to benefiting from all the work you've been doing behind the scenes
Also, just trying to learn how different people use grok bot in their day-to-day
Obviously, you have a really unique use case that we were talking about off-camera where you are applying to jobs
You said you need a job
And it seems like you've put together a pretty cool team of grok bots who are helping you do all that work
And so I would love for you to kind of just talk us through how are you using grok bot?
How has it been helpful for you in the job search and you know for those of us
Or for those of us watching maybe who aren't as familiar with like how a job search goes these days
Kind of like how that process works nowadays. I feel like it's a really competitive market particularly for software engineering
So maybe let's start off with that. You know, like how how does the job market work these days?
Like what are the things you're having to manage as you're kind of going through the job search process and let's start with that
Yeah, of course, so
Keep in mind that I started my job search my internship search around two or three months ago
Now there's usually like three to four pathways that I like to think of when it comes to the job search
First is obviously the normal method, which is cold applying, right?
You see that job for space x ai
Sweet internship and then you see that and then you're like, all right. I gotta apply within three hours
All right at the end of the day any big company is going to get over 200 500 applications
So it becomes very competitive very fast. The second approach is focusing on the startup community
all startups need
Good engineers at the end of the day and they need people who can ship fast deliver fast and then also bring in new
unique ideas that can improve like either their business or their workflow or anything else
The third approach is a little bit
A little bit unique in that sense, which is
Going going outside of the cold approach strategy and then just directly communicating with the recruiter
Right seeing if you can email them or having a coffee chat with some employee within
A particular organization you want to work for and then have them connected with the recruiter
And then get you fast-tracked into the process and then the fourth strategy is just like on air begging, you know
For example, I'm here as a forward deployed intern. So let's hope that stays
Let's hope that converts me
Because when you're mentioning that like there's so much that like this ties to when it comes to sales, right?
There's this sort of like you're doing some outbound prospecting where you're like going out and finding leads for your job
And then there's sort of the like the you know the open contact forms
Where you're filling in this like contact us, but you kind of want to make sure that goes somewhere
And so it feels like I can't help but draw that analogy and it feels like you're almost selling yourself
Right versus selling a product that sound right. Wait, exactly. That's a perfect analogy
Like you have to sell yourself when it comes to the job
Like whether it's cold applying you need to be the perfect match for
Whatever scanner they use when it comes to a startup if you're cool. If you're DMing them
You have to be the perfect candidate at the end of the day
You have craft like the perfect DM that like catches everyone's attention. Okay. Okay
Well, I know you're using grok bot to power a ton of these workflows
And so it'd be really great if we can just get your screen up on the camera. I think
Plugged in here. We can get that on the stream soon
Um through how you are using grok bot and I'm maybe like walk us through your team kind of who's in your team
There we go. If you don't mind just like command plus a little bit
I'll start off with the most recent one I talked about other than begging which is the recrure finder
So recordingאל Right now I have a grok bot that
Webdem on their vm it will
Find all different recruiters via search
Searches them up on-on uses the Apollo extension browser extension to see if they're their database
Copies that into a csv and then now now has to that
It goes through all the different big
big tech companies or spears that hiring for software
engineering roles. So for instance, you can see all the different rounds and all the different emails. Really hope this doesnn't go against any. Law about doxing but also. Also. As you as you a bunch of emails this. and this. Me see a
lot of different scripts you've written. Yes, cool notes,
you know,
really hope.
What this does is we have every university has like alumni database at the end of the day or a majority of them do and they all have information about which alumni are working where and for example I'm interested in like software engineering positions. So I'm looking into the alumni database and see who's working at the
Google, Microsoft, SpaceX, etc, etc. Then I search them up. So this bought what it does is it searches up a bunch of people runs it through the alumni database that alumni database hands me back the email and then it copies that puts into a CSV as per usual.
And then eventually what I'm going to do with both the the combination of both the recruiter finder and also the alumni database email finder is use that and then connect it with an outlook with the outlook mcp to just start sending out emails.
To these recruiters and alumni asking for coffee chats are asking for a way to understand the recruiting process right getting any information or any tips right all of those things are going to be very important during the job search and that is something I'm planning on doing.
Okay, so this I'm curious like how so you mentioned that there's like a database backing this how did you connect grok bot to that source of data so it's like a connector you build yourself to be to some CSVs like what does the process look like to kind of bring some of that context into the
box. Yeah, so I think the biggest thing is that there's actually no connectors actually really non technical in that sense I can honestly show you the screen.
Yeah, that's okay. Yeah, let's not let's not show any sensitive information. There's no sense.
But this is the alumni directory. It all it does is it just controls this web and it just searches up every different name every different profession, whatever whatever companies and then it finds all that information.
There's no connectors. Honestly, all you need to do with your grok bot is just give it access to whatever directory you want, and then also eventually give it access to your email that connects your grok bot to your student email, and then just start sending out emails.
It's actually very straightforward. There's not many connectors involved with this.
Cool. Yeah, I really like that. And it's a great example. I think of how grok bot can kind of give you its own has its own computer right to go get work done for you and it sounds like in this case you're just, you know, driving grok bot through computer use and having it just kind of navigate these various services
for you obviously something that we've seen like a ton of great use cases around over the last day or two.
Yeah, I'm curious. Okay, so we earlier saw a little flash in your finder window of cover letters. And so maybe tell us a little bit about your cover letter writer because we talked a lot about how this is a marketing and sales game.
Yeah.
So how are you marketing yourself?
Yeah, so that's actually a great question. And this is actually something I was working on five minutes ago, actually.
Well, slacking off while entering for us.
Sorry.
Sorry.
It was my downtime.
Okay, so the way that this works is I have two bots that actually work with cover letters.
One thing I've noticed with all the jobs that I've been applying for is you need two things you need a resume, obviously a tailored resume with the keywords or like whatever stacks and you're like implement based on the job description or the JD I will say JD a lot so that just
Second thing will be the cover letter which I think is the more the harder way to actually autonomize in that sense because cover letters are really personal it describes not only the job and what you've done but also needs to connect what you've done to the job include your personal experience
include your voice at the end of the day. And I think that is something that you just need to work on a lot in order to make that possible.
So what I have here is a set of two bots first is the cover letter writer.
This is actually a new bot that I created bus in like 10 minutes ago. What this does is it takes my resume it takes in all of the different stories that I've written within the spot saying, Oh, I work this internship at Stanford medicine.
Here are all the stories that relate to what I actually worked on. Right. Here's what I worked on for my project. This is why I worked on for my game right and these are all the stories.
It has all of that information. It also has my resume, obviously, then what it does is there's another bot that's like the sweet job applyer which focuses on finding all of the jobs out there.
GitHub repositories greenhouse. I forgot what it's called.
Yeah, exactly goes through all of that finds as many jobs as possible scans through the JD sends it to the cover letter writer.
Then it will take that information do a little bit of deep research up and needs to about like company values or whatnot.
Then it sends it then it sends it through a pipeline, which is just a cover letter creator repo that I personally made.
As you mentioned before, like off air, right, you saw cloud agents running. So this is like all the different cloud agents that were running before.
If I scroll up right here, what does that repo do we can get we can get your screen back on there in a second. Yeah, but what does that repo do is you said you wrote some code with a cloud agent.
Is this like sort of generating a PDF of the cover letters. Yeah, that's exactly it.
It generates the entire cover letter and it sends it back to the sweet job applyer.
The other thing that I wanted to work on because at the end of the day cover letters as I mentioned before really nuanced and any be perfectly crafted.
I created an actual like guardrail in that sense, like a cover letter critic, which looks at the cover layers and is like, Hmm, this doesn't sound like you.
I don't like how it's saying, and this is what I want to work on, which sounds so inspirationally uses. Yeah.
So that is something that I have right now, which is essentially like a guardrail at the end of the day, whether it's in cover letter generation or in like actual like code bases.
You need to have something that's writing the code, something that's actually thoroughly testing it and checking it.
Those are the two bots I have for the cover letter generation. This repo also creates the resume. However, I don't worry too much about the resume just because.
And this is also a tip that I would give to everyone who is applying to jobs. Do not fake your resume.
You will have experiences you have stuff that you can work on, or you've already worked on. Don't fake experience because when you get that first interview, which I have multiple times.
If you fake anything you are not.

===== END RUN 20260917-0554 =====

===== BEGIN RUN 20260917-0652 =====
of there. Right. So like we can start to like clean some of this stuff up. Like I don't want
shadows or highlights on these. Like generally I just want them to be simpler. I feel like it
started adding details. And I'd like more holistically make the card stack tighter.
So again, we've talked about design mode and cursor a little bit. I love the
like how tight the loop can be. And I can just kind of spam specifically models like four, six,
high. I mean, it could be low. Honestly, if we're just trying to move fast, but I can spam fast mode
and it'll just kind of like iterate on those things in real time. And I get this asset where
like essentially this is a video. And once I get this square video that looks the way I like it,
I can then just generate like portrait assets. I can generate really whatever I want. And
while a bit, you know, are you going to run an ad on this? Like probably not. We
should be running ads on this. But if you're someone who has access to these assets, and this
took us five minutes, it's not like hard to see how we could then, you know, build out this flow
where in GROC bot, we're just generating different kinds of remotion ads, right? Yeah.
I think it's also hints that like a really cool way to use agents. I think remotion is just
like magic. It's like such a magical tool. But I think the reason it's so magical with agents
is that it speaks code, right? So it takes this like what you, you know, what is still
really a laborious process to animate and like, you know, go into after effects. And
there are people who are like really, really good at this work, but it provides a coding interface
on top of it. And I, every time I see someone use a motion, I'm always a little bit blown away
with an agent. And it always sort of reminds me like, you sort of like got to play to the
agent's strengths sometimes too, right? Like you earlier, we're talking about PDF formatting
and text formatting. Like obviously agents are really good at text. And so kind of,
you can kind of lean into that to get some leverage out of them. I think similarly,
like agents are really good at code. Like we've been talking this entire time about how, you know,
like coding is cheap and it's really easy to produce code. If you can take the act of creating
graphics and motion graphics and make that, like put a code, put a code interface in front of that,
I think you're sort of like speaking the agent's language in some ways. And so,
you know, one sort of pro tip for folks who are watching and following along is, you know,
think of the tools that you're using and how you could try to, you know, make code, make
them code more, more code enabled or build sort of harnesses that involve programming on top
of them. And I think that actually helps your agent school a lot further, both in terms of what
they can do, but also just in terms of like, you know, tokens they burn and the efficiency and
all that kind of stuff too. So, 100%. Yeah. And I mean, I have this like grand vision where
everything is code, like in docs, you don't have screenshots. You just have code. The XAI and
the cursor teams are already pretty good about this, but you could imagine we change the
component in the product and it just flows through our documentation or our content. Yeah.
That would make me happy. Oh man. Okay, cool. So we got some hats cooking.
I feel like we made some progress there. Thanks for joining us. This has been awesome.
Really appreciate you kind of hanging out with us. Are you on X if people want to find
you afterwards? Yeah, I'm on X at Chardul Marate, I think all lowercase. Nice. All right.
All right. We're doing it. We're doing it actually before you get out of here. Chard.
Oh, S. Sorry. I'm S-H-A-R-D-L, right? Yeah. M-A-R. A-T-A-G. A-T-A-G.
All right. This is for anyone who's a recruiter who's watching.
Chardul is awesome. He gets the follow-up, by the way.
And then we're going to do a quick, we're going to do a quick profile breakdown
because I think, you know, it was actually, I'm surprised we made it this far on the
stream without breaking down X profile. Notice how locked in Chardul is here.
Like, I just want you all to see my man is like, we're going to,
I'm going to, we're going to clean shot and then we're going to edit this.
Gaze is directly at the computer. Like, we get arrow, eyes. There's nothing,
there's none of this. There's not looking this way. My man both lays their vision on the machine.
So that's one. We're not going to save this image. Maybe for later. We'll keep it on my desktop.
So first one, locked in check. That's A-plus. Second, this is a great head shot. I feel
like people, this is a good looking guy. A lot of people that, you know,
they do like the, the, the A non-PFP situation. I like that you're out here. You're repping the name.
You're being just very out there with it. A-plus there. We've got the portfolio site.
We're going to do a quick portfolio. I love my Yu-Gi. I love Yu-Gi. I will.
Chardul, I'm so sorry we brought you on here and then I just started going through your stuff.
Hey, no, I'm proud of my website. I'm proud of my website. I will say that.
It's a trap card. There's some good stuff there though. He's like leveraging that hero photo from
the core. Yes. Some clear speaking engagements. Some more head shots. Actually, the glow.
Lauren, we should take some inspiration from him. This is so many, so many stars. Okay. A-plus
portfolio site. You should click on the projects. See what animation. Okay. Okay. Okay.
I don't know if we have enough time to go into all this guys. Production is telling us
that I'm doing too much. A-plus Twitter profile. You got the follow. It was great to have you on.
Thank you so much. Thank you. We're going to run a quick promo for those who haven't
followed along yet. We are giving away a chance to win a visit to Starbase, Texas. So if we
can cut to my screen for a minute, we'll show that very briefly on stream and then we will
switch over to our next talk downstairs in the main theater. So if we can get my screen up,
please. All right. So Starbase, just share your best Grockbot templates with us, this QR code.
We can move our camera off of the screen for a second. We'll be able to see the QR code
hiding underneath. There you go. If you want to win a trip to Starbase, Texas to see a launch
of Starship for you and A-plus One, you can scan this QR code to learn how to enter. You've
got to share your favorite Grockbot templates with us. And then we will have several runners
up who will win a trip to learn to see where some of the stuff is built that powers some of
these incredible, incredible vehicles. So be sure to enter this. If you missed the QR code,
when we flip it off here, you could just follow at Grock on X and we have some instructions
on some of the latest posts on that account too. Share by September 29th and we'd love to see
some really great templates. With that, I think we're going to cut downstairs to the
main stage for our next talk. I'm Simon. I'm here on SpaceX AI's Go To Market team,
specifically in SDR. Excited to talk about Grockbot specifically for this team.
At a high level, we'll be going over what actually is Grockbot, why we use it internally,
and then dive into some of the actual SDR use cases we use. Then I'll demo what my
environment looks like and then we can discuss that and go into a Q&A from there.
First, at a high level, I wanted to discuss the actual AI maturity curve
and what this looks like for SDRs. Starting out, when I was using AI,
I was just prompting back and forth with a chatbot. I do typical things like I'd ask,
can you help me actually draft this email or perfect this copy or do this account research
for this specific account I'm looking at? What we've seen progress over the last year
is first idea of co-pilots. Co-pilots are AI that can actually go out and carry out a true
task. Instead of talking back and forth, getting email copied and then copying that into your Gmail,
it's something now that we see actually having the AI go and basically hook up to a Gmail,
MCP, or an API and be able to actually send that email off for you. Really with bot, what we
believe is the next frontier is actually automating entire jobs in full. Going beyond just a task
and actually doing an end-to-end workflow. Really with the staff function of actually
combining a team of bots, you can see how kind of automating a job, but also combining that with
different features of business, like talking to RevOps or talking to your AI and getting
contacts on an account and how bots can actually pull in that context for you and really help you
do that full process. Just diving into what Grockbot actually looks like if you haven't seen it
before. You have all your bots kind of here on the left-hand side and they're doing
different things like we have a sales outbound bot here, we have our chief of staff and
you can message your own teammates. More interestingly, you can actually see them
talk with each other. I have my own Grockbot instance set up where really there's only one bot
that I'm actually talking to. For the most part, I've kind of offloaded all that context
to my chief of staff actually talking with other bots and doing this. I think some
important things also to emphasize here is that bots keep their own context at a per bot
level and memory so they can really, improve
as you work more with them and as you go, their memory gets better.
This is why I really emphasize you actually should be that training your bots.
I'll really go into more of what that actually looks like.
I think the truth, at least it's face sex of what we've seen,
is I message or anotherTC. And I think more important bots are always on. And so you don't
have to walk around with your laptop
to be like, okay, there's no real open API for it. And so this is really something that I found
immense value in. And what's given me the confidence to be able to say, okay, Grockbot, you can
genuinely go end to end with what you're doing here. And as I talked about before, I
think shareable templates are one of the most important things. And I think every single
team at a team level should basically all be spending time building their own templates,
sharing those with the team, getting understanding of, okay, what are the things
that are actually working? And how can I not only just import that template,
but how can I import it and then add my own nuance onto that of whatever my tonality and my messaging is,
or actually, maybe it's just I like to do my workflow a slightly different way. But actually
kind of taking this as a base template and then working beyond that. And you can also
prompt Grockbot directly and ask for direct improvements. You can import templates and be
like, hey, look at this template, my colleague shared with me, what do you think are things with
of my own workflow that you think would be a good addition
to build on top of this,
or to kind of get more value out of that.
So I really emphasize just actually like talking back
and forth with your GrockBots
and kind of like starting to really push it to help you.
And I think this is why we emphasize so heavily
that GrockBot is really just a multiplier for your own work
and really way to scale the outputs you're doing.
And as an SDR,
I don't think there's anything more important
than actually doing that.
All right, so diving into really what I see
as the common SDR use cases
are the things that like I'm doing on a day-to-day basis.
I think number one, it's really prospecting.
And so if you're an early-stage startup,
you can actually even help have GrockBot
kind of help you understand your ICT.
And so you can look into things like,
okay, for the AEs that have closed deals
or have deals that have moved to stage one,
what is the common kind of title
that we're seeing actually carry those deals through?
Or what's kind of the most responsive title
or industry or sub-industry of person
to take an original meeting?
How does that actually vary from economic buyer,
, the person who's actually signing off
on this deal?
And so you can really have GrockBot
kind of from a first beginning actually understand
like what is my ICT?
What's the most optimal way to reach my goal?
And personally as an SDR for me,
that's how many S1s can I get?
And so you can kind of reverse engineer from that point
and GrockBot can kind of help you
actually get to that point.
And so on top of really just understanding that ICT
within prospecting, you can do the full flow.
So you can use enrichment tools,
set those up within GrockBot.
You can also use web search to kind of find
more specific data like job postings
or any funding announcements
from some of your target companies.
And really just bring this all in.
And I'll show how I can actually tie that
into my next point, which is a sequencer.
And so really what I've done
within my own GrockBot instance
is I built kind of an ongoing list
of all my prospects.
And this lives as a CSV today,
but really what I think is nuanced about this
is you can have things like GrockBot routines
and automations, which are constantly updating this.
And I think something that's really been a big value add
on the sense of sequencing,
is the fact that when I'm multi-threading an account,
my AED might have a call with them
and that might have happened yesterday.
And whatever the context from that call is,
is truly golden and it should shape my messaging for today.
And so what GrockBot can do
is it can actually take that context in
just like a coworker would,
they can go, hey, your AED spoke
with this person yesterday.
Here's exactly what they mentioned.
They mentioned a real pain-point is
they have all these tools to choose from
and they kind of just feel stressed out.
There's tool fatigue, there's AI fatigue.
And really-
And really that should shape your messaging
for that day going forward for that company.
And GrockBot has that context
and it can pull that in
and help you actually tailor messages
at a company level.
And maybe even in a Gong Caller recording,
they mentioned, oh, you know what?
Like this is actually the exact thing
that Simon on the AI team
is investigating right now or looking into.
You should reach out to him.
But a lot of those things never actually get action on.
And so GrockBot has the ability
to actually really sequence and kind of
do the full process of prospecting them,
adding them into the sequence,
just off of the actual context
that meeting they had with your AED.
And I think that's something really nuanced
that I haven't seen before.
And on that same note,
you can also do very high level account research.
And so for PLG-led companies,
you can look at all the internal data of,
okay, who are my power users?
Who's actually using this product?
And more than that,
look at the external signals for all companies in whole.
So okay, who's fundraising?
Who has job postings?
And how does that kind of table the messaging
based on our product?
You can also look at things internally,
like okay, how does this new product that we've launched
actually change our outbounding message
for a specific person in XYZ industry?
And really have GrockBot kind of table those messaging
to each and every company specific.
And that's actually diving into my last point as well,
which is drafting copy and what that looks like.
And so what I really highly emphasize
is you actually sync your email with GrockBot,
just as kind of a starting point,
and really push GrockBot to understand your tone.
And I think some of the nuances that I've added on this,
because it's very easy to just sync your email,
but you have to kind of push it further.
And the first thing that I've done on that note
is I said, listen,
look specifically at my email copy
that I've sent out to external or third party companies
that also match the Salesforce account
where I'm an assigned SDR.
And then from that point,
what I actually am pushing it to do,
it's also look at, okay, just look at any of the emails
from that filtered list
that actually got positive responses,
that didn't get negative responses.
And then I think the important thing that I've done
on top of that note
is I've actually put a heavier weight on emails
that recently have received positive responses.
And I think for a company like SpaceX,
what we're always seeing is
the product offerings are always changing.
Like I said, what was working three months ago
is completely different now and might not work today.
And so I pushed Grockbach to put a really heavy emphasis
on the actual recent messages
and to draft my copy more like that,
but still kind of pull in inspiration
from my older messaging
to understand like kind of the human aspect
of actually what I'm doing.
And so these are kind of the real SDR use cases
that I've seen,
but I'll show you more in my demo
of kind of how we can expand much beyond this.
So at a high level, I just wanted to first show
what does my Grockbach team actually look like here?
And so what I've really been emphasizing
is you should have a chief of staff.
And I think one of the most exhausting things for me
is switching back and forth between my different bots.
I'm always kind of forgetting the context,
especially if I'm kind of shuffling
many, many things at once.
And so what I actually really recommend
is you have a chief of staff
and that should be the first bot that you set up.
And what you should actually do from that point
is set up all your other bots through your chief of staff.
And the reason being is when you do it this way,
your chief of staff just has context
on exactly the purpose and the end goal of that bot.
And so as you kind of ask your chief of staff,
so Simon bought here for me,
to complete tasks,
it understands the right bot to actually delegate that to.
And when needed, it's gonna understand the context
of multiple bots that it needs to delegate to
or actually get to talk with each other.
And so I think this is an important way
to actually think from the beginning of how to set this up.
And just for my other bots here, you can see,
so I've got Shakespeare here,
who is really the one I've very meticulously trained
on my email writing skills.
And then I have web search,
which obviously is just doing web research,
but he actually works in parallel with my Simon soldier.
Now I'll show you, I basically have an entire army
of Simon soldiers,
which I basically consider sub agents or sub bots
that when web search wants to delegate a task
across say 50 or 100 counts,
the scalable way to do that is actually creating sub bots
that basically respond directly to web search.
And lastly, I have customer bot who does a few things,
but really is just taking in all the context
on the meetings I'm having,
understanding for deals that are moving to close one,
kind of what's the context,
what are the things that customers
are really receptive to.
And you can also get really nuanced with this.
So I think something you can do is look in Salesforce
and say, okay, I have an account
that's in the FinTech vertical.
Okay, let me understand,
let me talk with customer bot
or ask my chief of staff to actually talk to the customer bot
and figure out, okay, what are the similar FinTech companies
that are closed one in recent deals
and why were they interested?
What were the main pain points they had?
What were the main parts of a product
that were actually relevant?
And then I'm gonna take that context
and I'm actually gonna have my chief of staff
pass that off to Shakespeare
to understand my email drafting
and to push that in specifically
for the FinTech companies we serve.
And so this is where I really pushed
that you can have these all as kind of your entire staff
working on different parts of your outbound
to make it genuinely productive.
All right, so now let's dive into the demo
and see actually what my setup looks like here.
So here we've got it at a high level.
I'm very serious about my color coding
as you can see here,
but I promise I have a reason behind this.
And that's, I really like to understand at a high level
when my bots are talking to each other
and especially because all my work is delegated
from my actual chief of staff.
I need to understand who my chief of staff
is talking to at a section level.
And so whenever I see the orange here,
I know that my chief of staff
is working on an actual research mission.
It's doing something around research.
When I see that it's message Shakespeare here,
I know that it's doing some sort of outbound copy drafting
and it's kind of starting to understand that.
And the same here, I've talked about my army here,
but when I see the green messages,
I understand, okay, it's delegating account research,
it's splitting these up between subbots.
And you can actually see here,
I have an army huddle as well,
which basically has my web search bot
talking with all of the soldiers here
and basically just ordering them around.
And the soldiers are just,
they're giving all the context they need.
And so you can see I'm researching
kind of this demo account Fly Low here,
but you can see how it's basically broken all this up.
And so now, oh yeah, yeah, for sure.
Good, all right.
So now what I really wanna show is actually
how I'm enriching prospects,
how I'm actually going and sequencing these people,
and also what my daily routines look like.
Because I think personally as an SDR,
one of the most important things for me
is waking up every morning and knowing very much
what I'm actually going to work on for that day.
I think that's just having a signed task
that you wake up to is extremely helpful to start with.
And so really what I've done here
on my routines on the side here.
So the number one most important thing for me
is every day I want 50 new prospects.
And really these prospects have their own
kind of internal rankings that I've trained
graphite on.
And so typically how I break this up
is I have five prospects that I need to action on
first thing in the morning.
These are gonna be the most relevant ones.
So maybe it was someone who contacted sales,
but nobody ever reached out to them.
Maybe it was someone who just downloaded
some of our content that's extremely relevant
to what they're doing in conversations
we're already having with that account.
And basically this is my first five people
that I genuinely need to action on right away.
And what I have is you can actually set up Grockbot
to things like your calendar.
So my chief of staff has access to my calendar.
And I can basically understand the context of,
this is the five things you need to action on now.
You have a 15 minute block I can make for you
during this time, maybe 9 a.m.
So you can actually just knock those out of the way then
you don't have any other meetings.
For the 45 others that I know you wanna get to today
I'll block off time at 1 p.m.
So you can actually carry those out over another hour.
But really my chief of staff has kind of started
to understand how I actually like to kind of
break up my time and what's most important
and really how I've seen the most results here.
And so on this same basis I have an inbox manager
who does something very similar.
So I get Slack messages that I wake up to every morning.
I'm added to meetings when I wake up every morning
and I have external emails that I wake up to every morning.
And I really just need a simple way to actually understand
which of these I should action on right away,
which of these are most important.
And so with a similar system this is actually just going
and it's really just ranking these for me
on ones to action on.
And what the order to actually do that is.
And so on this on the same basis as a PLG company
something that's really important to us is actually looking
through all of our accounts for some of the signals.
So where are power users living?
What are the products they're using?
What are the things that are actually using the products
for that are most relevant?
And so I also have every single weekday
and you can see all these are at 8 a.m.
I have all of my accounts basically getting scanned
for any net new signals.
So who are the new people who actually signed up
for our product?
Who are the new people who have reached out
to download the content like everything?
And I really like this because I think speed to lead
is one of the most important things.
So even if it's not the strongest signal
being fed that on a daily basis
so you can actually just go in action on that
while it's as warm as possible has led me
to some of the most receptive responses I've gotten.
And so finishing out here I have my sequencer
and I think just an important nuance on my sequencer
is that I've also have a routine
basically set on a Salesforce trigger.
So anytime one of my accounts,
the actual account status updates.
So maybe it's moving from stage zero to stage one.
This is basically a trigger that works in parallel
with the sequence to unsequence people
when maybe the deal is moving forward
and we don't need to outbound anymore people.
And so this is really a way you can kind of set up
these kind of elaborate systems
all just with routines and with Graphbot itself.
But I'm gonna dive in now
to what my actual sequencer looks like here.
And so here's kind of what it looks like.
I've spun up this demo one here.
And so you can see it's all saved as a CSV
which I will admit is like a hideous UI
and it's like very painful to look at as a user.
And I think what my big emphasis here is
you actually as a user shouldn't be looking at this much
if at all.
This should kind of just be Graphbot's brain
and some systematic way to actually track these.
But you should genuinely trust Graphbot
to actually have the context and only look into it
when you actually really need to.
And so you can see here
I have all these different sequences,
a queue of prospects
where I actually need to be at in my prospect board.
And then the most important stuff for me
is the things I need to action on this specific day.
And so you can see this basically gives me all the things.
It's also drafting copy for the emails and everything.
You can see here it's already got Gmail Drafts ID.
So I can basically just go into my Gmail
and review these all one by one.
And if I wanted to, I can just talk to my chief of staff
and I can be like, can you actually give me a score
on how confident are you that this email is accurate,
that it's solid
and that they might actually give a response.
And this is kind of what I've delegated it to
at this point.
I think I really emphasize that first
and especially as you're kind of training it on your voice
to actually like thoroughly look at these
and honestly give it feedback.
But it's now at a point where I somewhat just trust it
and I can actually push it
to give me its own rating scores on how good these are.
But I think on that note something important to emphasize
when you're focusing on your copy.
So this could be Gmail,
this could be for any other platform
is what I've found works best
is actually just prompting back and forth with Grockbot
and asking it for examples.
And so honestly when I first had it pull in my emails
and I even did all the stuff, okay,
like let's wait the more recent responses.
I found that it was almost kind of template
and every response looked very similar,
but just with swapping of names of company IDs,
things like that.
And so what I've really done is just kept going back
and forth and asking, can you give me examples
of these emails?
And then on each one basically critiquing it
and being like, this is why this email sucks
and here's how you can make it better.
And really pushing for it to break out
of kind of the standardized formatting it has
and to actually kind of think
and give nuanced responses at each single email.
And I think something I really push
is not a single one of your emails
should look like a template.
It shouldn't look like the same thing
but just swapping out company or name.
And I think that's something really important
to also push here.
So I'm gonna dive now into actually like
why I structured my setup this way
and what this actually looks like.
And so obviously I'm my chief of staff at the top.
And as I said, this is actually the only bot I talked to
and the one that I built the rest of my bots from.
And so obviously I have all my outbound
actually just structured in a single folder.
And the reason I do this is I make one bot
basically per platform.
And I think the important reason to do this
is because you should be thinking about bots
as a way of things that can run in parallel.
And so Shakespeare is doing all my email.
And so I don't need a second email bot
but I might need a second outbound bot
for a second platform that I'm reaching out to people on.
Or maybe I need context for some of the calls I'm doing.
And so having different bots
for each of those different functions
I think is the best way to actually handle this
and understand the context
of how your bots are working with each other.
And then so diving into my research here
so kind of the six key bots
that I'm actually using for this.
One, I have my PLG bot which is tied up to Salesforce.
So I'm actually getting to see,
okay, who's signing up?
What does our usage look like?
It's also pulling in stuff like,
do we have any closed loss?
It's also talking with other bots.
And so chief of staff might be able to recognize
that, okay, one of our prospect accounts
is a closed loss opportunity.
Okay, what's the context on why?
Well, let's dig in the voice of the customer bot,
understand the reasons why that opportunity.

===== END RUN 20260917-0652 =====

===== BEGIN RUN 20260917-0725 =====
And it can talk with other bots,
like our company research bot here,
pull that context back in of who exactly that is,
where they sit in the work chart,
and help you actually sequence that
and rank that based on here's how high
they up in the company they are,
like are they the economic buyer based on stuff like that.
And so really you can kind of have these all work together.
And again, I think this goes back to the point of
the importance of building all your bots
from your chief of staff bot is
it understands this context better.
And it understands it just like you would.
And so moving forward here,
I've kind of talked a lot about the voice the customer
updates on summits.
And so how do we actually like scan for this at scale?
So I set up my web search bot here,
which is hooked up to Exa.
And what I actually do have is an army for it,
I can actually break up and delegate this work.
And so you can see I kind of talked about this earlier,
but I have an army huddle,
which is basically just one bot telling the other bots,
okay, we're gonna break up these 200 companies to 40 each,
now run in parallel to actually do web search
on all of these.
So we can actually get this information
back to Simon in a timely manner.
And this is something I've really seen as like a practical way
to scale some of the long running tasks
and some of the ones that really have slowed down
some of my workflows in the short term.
And so this has really helped me kind of stay
on top of those things.
And so kind of on top of that,
also we're looking at usage for PLG companies,
we're looking into how much are people using this product,
what teams are using this product.
And I think for Gawkbot it's something important,
we're trying to understand like
how many go to market people are using it
and what are kind of the use cases they have.
So being able to pull in data like that too
and have my chief of staff actually understand
all that context makes it all the more easy
to actually like kind of rank every single prospect,
understand priorities and understand really like
who are the best targets to go after?
Cause I think it's so easy to be overwhelmed
as an SDR with all these different tasks
and all these different things you have to do.
And it's really hard to kind of mentally
try and understand and rank these different prospects.
And so I think having some sort of like systematic ranking
here with Gawkbot as I've kind of built myself
is a really good way to do this.
And like I said earlier,
I think a really nice part of this
is you can actually share these with your team.
And so you can get the learnings
and kind of share those learnings.
And I think one last thing I just wanted to show here
is I've actually built some skills
that kind of go deeper here.
And so I actually have a entire just prospecting skill.
And so our company here is Flylow,
but we can just type this in.
And so prospecting understands already
all of the different things I want.
And this is more something at this point
I just run for one off maybe net new accounts I have
or things I wanna research like midway the day.
But most of this runs at routines at this point.
But I think something very interesting
that I also wanna talk about
is the skill I've made around Gawkbot ICP.
And I think this is something
that's really widely applicable here.
And what this is is we're still figuring out
all the different use cases for Gawkbot.
And we're not sure really like who the best ICP
and what that looks like.
And it's really something that's kind of
an ongoing development.
And so what I wanted to do here
is I wanted to turn our own findings on Gawkbot's ICP
into a skill.
And I did this because this way
I can constantly update this
as it kind of gets more and more clear
what this actually means.
And so obviously it's something I've talked with Gawkbot
before about okay, let's see.
What are the deals for Gawkbot
that are actually moving past stage one
that are showing interest?
What are the things from Voice of the Customer Bot
that are showing interest around this?
And why are they interested in this?
And this has really helped me kind of
iterate on the skill more and more.
And I think it's something
that really I wanna keep building out.
And the reason I've kept it as a skill though
is because it could drastically change.
And I want it to be able to be flexible
and be something that my systems
can basically call upon at any point.
And it's updated.
So as we kind of iron out more and more
what Gawkbot's ICP is,
we can really change this on the fly
if we save it as a skill here.
And so basically I can run that in the background
and you can kind of see
some of the high level stuff it's gonna do.
But really I do all my work
and I kind of orient it all around CSVs
that look like this,
that are pulling in everything like titles.
Like what's their actual ICP persona?
Like the owners and everything.
Like what's the context on why?
And then you can see it's talking to other sources
like Ample Market to actually get email data,
LinkedIn, locations here.
And really just a lot of the context on these profiles.
And you can see that it's already basically running
and doing something like this.
But really at a high level
this is how I actually run all of my sequencing here.
And you can see that now it's messaging
all of my orange bots.
And I think just for my own sanity,
it's very clear now that it's in the research phase.
And I understand that.
I think now, boom, it's doing web search in the background.
And I'm very clear in understanding
on exactly what it's doing there.
And that's I think why it's important
to kind of actually do these somewhat
in a color coded fashion or maybe some themes.
It's easier to remember
and to actually kind of at a high level look down
and see what's happening.
But all right, that is the demo.
And so now I'm kind of just gonna discuss
like what we actually learn
what I think are genuinely like the takeaways
that you guys should have and carry into your own setups
for this is SCRs.
All right, so I think one of the most important things
is going end to end.
And this is something I hope I showed in my demo
but then I'm really trying to push
is that it's very simple to just like prompt back
and forth with GrockBot
or to have GrockBot do a single task
like sending an email for you.
But I think really you should push it
to do the entire workflow.
And it's a lot of pain upfront
for you to prompt back and forth
to build this system to figure out
like what actually makes this system work.
But if you stick through it and you're persistent
and you train it on your style,
you will get so much more benefit
when it's able to go end to end,
when it's able to pull people in from prospecting,
it's able to enrich them,
it's able to understand all the context
and then it's able to push that out
and actually outbound them and do the full process.
And I think that's really what differentiates GrockBot
and what makes it a true like force multiplier for SCRs.
So I really emphasize like thinking at a high level
and kind of thinking how you can actually build that system.
And I think on that same note,
you should be intentional here.
So it's really intuitive to just prompt each bot
and what I highly recommend is just taking a step back
and thinking, okay, let me build this all
through my actual chief of staff.
Bots should be able to talk to each other.
They should have that context.
The point of this is to offload more work from you
and help you actually get higher outputs.
And so you shouldn't actually be focusing even harder
and messaging even more bots.
It's exhausting from just all the prompt
and context switching you're doing there.
And lastly, what I kind of hope I really showed in my demo
is how to think systematically about your setup.
I think it is very easy to just always want
to make new bots.
It's kind of fun and it's kind of exciting.
But as someone who's built an army of bots,
I promise you I've had way too many at some points
and it's honestly more chaotic
and it does more harm than good.
And so you should really be questioning like,
why isn't it important that a new bot does this?
Why can my existing email bot not just write my emails
and also be able to be my inbox manager?
Like, why do I need a second bot for that?
And also on that note,
when should I just turn things into routines
that that bot runs?
But really I emphasize kind of thinking
about all these different things
as ways for you to actually just save more tokens
in the first place,
but actually just also not confuse yourself
and not kind of just get overloaded
with bots that are just doing kind of similar things.
And again, making these through your chief of staff
makes it all the more easier to actually do that.
All right, so I just wanted to thank you guys
for taking the time
and I think we're gonna just open it up to questions now here.
Go for it.
Yeah, so you had a daily routine
for 50 prospects and 50 leads, right?
So if you keep running the same workflow every day,
approximately what sort of consumption
from a token perspective do you see, right, for this?
And do you have any numbers that you could share
like per day, per week kind of thing?
Like what that translates to into a dollar amount
or anything like that?
Yeah, so the question is,
how many tokens does it kind of consume
to run these on a daily basis?
And I think it very much depends and it's hard to say.
And I think the best way to actually do it
is you kind of have to see what it looks like
for your own instance because it depends on kind of
the amount of maybe gong calls you're pulling from,
the amount of context you have in your Databricks instance.
And also on top of that,
like how many prospects do you need to freshly enrich
versus how many are kind of old contexts
that you're pulling in from Salesforce
that you already have context on?
So I think that it very much depends.
It also depends at the scale you're running this on.
But I think the best way to kind of do this
is honestly like prompt back and forth with Grokbot,
get an understanding of where your token usage is at.
And you can also talk back and forth with Grokbot
to find ways to optimize your workflow.
So maybe 50 daily prospects
isn't something practical,
depending on like the scale of outbound you're doing.
Like maybe it makes more sense
to actually run 250 on a weekly basis.
And that just sends on a Monday morning.
And then you have that to work through the whole week.
I think mine might be just a little more daily
because I like pulling in all that context
and really like pushing Grokbot as hard as I can.
And I think I'm just using a little more tokens doing that.
But yeah, I highly recommend just trying it out
on that front.
Yeah, go for it.
Yeah, first of all,
thank you so much for a great demo and the presentation.
It's really fascinating to see the actual,
the use case from your own workflow.
So I think the part of your demo,
I think I saw,
you kind of have a group chat among your boards.
So I'm curious, like where do you see the benefit
of creating that kind of the group chat
because you already have a chief of staff
already connected all of your boards.
So what is the use case that you found
like it's more beneficial?
And my second question is,
I really like the naming of the Simon soldier.
So the other like the orange boats
were pretty intuitive for me
regarding what they're like the role each of them is.
But I'm just curious,
what is the soldier boat for you
and then how do you use it for your workflow?
Yeah, so I built my soldiers really just as like,
okay, like these are my troopers
that I'm gonna send out
to go do web research in general.
And I think the reason I put them in a group chat
instead of actually like setting them up
just as messaging my chief of staff directly
is one, I think it's easier to see the outputs
come back to my actual web search bot.
So I can actually go in there
and make sure that outputs at each of the sub agents
is somewhat accurate and is flowing correctly.
And I think on top of that,
it's much easier if I wanted to scale that
from five people in my army to 20 or something.
It's much easier to scale that
if I have my web search bot
basically handing off the context.
And all those sub agents really have like low context.
And so it's very easy just in that way
to scale that up or down on a whim.
And so that's kind of like why I thought
of doing it that way.
Sorry, I might've missed it, but like I guess like,
so if you, let's say you come in, right?
And you create a chief of staff bot,
can you then just task the chief of,
like just show what like your work is
or what you need to be doing.
And then the chief of staff like spins up
the appropriate number of bots automatically.
And where do these sub agents kind of come in?
I guess like, because I think when I come in,
sorry, you were saying like,
you should think about work being done in parallel
and shows there should be different bots
for things being done in parallel.
But that's difficult for like a user perhaps
to think about unless I guess maybe
you just really, the know that job very well,
but it could be different every time.
But what's like the way to just get going really quickly
where the time to value is minimized
as someone just comes in,
just wants to set up for me based on best practices.
Perfect, I'll answer this quickly.
And then I think we're gonna go back
to live stream upstairs.
But I think at a high level,
really just talking back and forth with Grokbot
when some of this stuff is extremely overwhelming,
I think Grokbot can actually help you understand
what a practical setup is.
And if you just kind of talk it at a high level
of what your actual workflow looks like
and what that is, it can really help you kind of divide up
when it makes sense to use new bots.
And I think that's kind of the easiest way
to kind of answer that broadly
for all the different use cases.
Just one quick, they hand me the mic.
First off, I love the Simon Soldiers.
I was telling everyone at dinner last night
about my beautiful, beautiful, sweet Grokbots.
And I call them all Soldiers for Christ.
Oh, right on.
Because they march forward on my behalf.
All right, welcome back.
We are back here in our studio back on the stream,
back with some updates on what we've been building.
If we can get my screen up on the slides
or on the stream here,
wanted to make sure you all could see that slide
that we flashed up at the end of that talk.
So you could scan that QR code if you haven't already.
While we get my screen up,
if you want to win some free Grokbot, you can.
Follow these steps to get a free month of Grokbot
for $200 in value.
So it's a pretty big free month at our highest tier.
If you duplicate and create a bot with Dr. Eggbot,
so go get the Dr. Eggbot template.
Create a bot with Dr. Eggbot.
And go through the experience of installing
Dr. Eggbot into your own Grokbot instance.
If you do that on stream,
following this QR code,
the first 1,000 users who go through this
will get a free month of Grokbot
to be able to do all the things that we've been doing
and much, much more $200 in value.
It's a lot of credits.
It's a lot of usage.
And we're really excited to give you all a chance
to kind of do a little bit of what we've been doing.
If you haven't had a chance to try Grokbot already,
great way to get your feet wet with a lot of credits.
And it's an exclusive on-stream offer.
So now's your chance.
Grab it before it's gone.
Okay, well, welcome back.
We are working on our business here.
We've been spending the last few days
working on building a gaming business and a game studio.
And for those who haven't been following along
since, I don't know how many hours it's been,
we are some number of hours.
Yeah, it's like almost over two days into this process.
But the plans are starting to come together.
We've had a lot of fun guests on today.
A few of us have been peeling off
to go work on some of the builds in the background too.
And so we've been really excited to make some progress,
show you guys what we've been up to.
I've got a few bots that we're gonna cook up
and kind of walk you some demos on today.
Where do we leave things off?
So I think what I did while during our last break
was I spent some time taking the prototype
that we built and turning it into a Next.js React app
with some placeholders for the server side
that we're gonna build.
So we're gonna kind of split out the client
and the server app into separate apps
so that we can evolve those things independently.
And Lauren, you were sharing a little bit about how
this is gonna make it much easier
for us to rev on the client, right?
Yeah, I mean, I think for the client
and I mean, for like a competitive game like this,
you do want to have a server that is deciding
like the things for like people,
like you can just modify the client and cheat.
So I think that will be a good thing.
We'll have off as well set up.
And I'm still not 100% sure
on what the client will look like.
I think like this, what we have
with the web version looks fine so far.
But it makes me wonder if like maybe we should,
it'd be interesting to get a few bots,
you know, started up maybe with Dr. Eggbot.
And then maybe even prototyping
different versions of the client.
So now that we have a backend,
like we can have multiple clients.
Well, I don't think we'll have multiple clients at launch,
but for us to iterate,
because maybe there's an interesting 3D version
or I don't know, a pixel art version.
Get the Blender MCP involved.
Terminal, terminal, client.
Yeah, yeah.
Well, what's cool about this is,
now that we've kind of separated these things,
we can have the server maintain the state
and then just like follow the server contract
that we've established with that API
and go and let the client, you know,
prototype a little bit.
We've added in a few new features.
We have a global leaderboard here.
It's obviously not fully wired in.
So these are all just fun names for now,
but hopefully be able to get that.
So once you guys are into the game,
we're gonna have a leaderboard and then I'm actually excited
to get something launched in the next few hours.
So what I was thinking is maybe to help us
make some progress is let's start kind of tracking
all the things we need to do.
So I have my chief of staff bot here
and I'm gonna ask my chief of staff bot
to get our Notion.going that has our to-do list
and then help us keep an eye on it
so we can track all of our ongoing tasks together.
We've all been dumping some stuff into Notion.
I think as we go, Lauren, I think you have a great,
a game design bot.
Oh yeah, I spent up a game designer bot
to do some research on.
Well, I think like our game mechanics
are pretty simple right now.
So I created, I got Dr. Eggbot to make me a new bot
to do, I guess, game research.
Cool.
So it looked at some other battler games
for interesting mechanics and interesting features
for us to be inspired by.
And I got it to write a Notion doc
that's not wired up to anything right now.
It's just a standalone doc,
but it'd be interesting to have like a,
you know, as we start to build out the game studio business,
start just thinking about like operationally,
you know, like how do we actually keep each other informed
of what we're doing?
And like, it could be nice to have a Notion doc
that's just tracking all of the work
we're doing on the project.
We don't really, I guess we do have a deadline.
All right, we have some milestones we wanna hit
so it'd be good to actually start tracking that.
Yeah, let's do it.
I have my chief over here working on that,
and then we have my founding engineer,
Bake, who's been keeping an eye on all the PRs
that have been coming into the repo.
So we have a bunch of PRs that we've been shipping
and I kind of have this, my bot over here is,
well, let's see if I can pull this up the right way,
is yeah, watching the repo.
So you can see there's some GitHub,
kind of, you know, integrations tied in here
where anytime a new PR is opened or changed,
I'm getting notified about it in my bot.
So that's been really helpful for me to see
on top of all the changes Lauren's been making.
We've been dual merging here.
So I'm kind of making changes to the client.
Lauren's been making changes to the server and backend.
And so you can see I got some updates here, which is great.
And then my bots are messaging each other
to put together this plan.
So as soon as we have that,
I'll pop up a Notion page and hopefully we can
get that going and look at it together.
Maybe just like pin it to the chat over,
pin it to the stream over here.
So we have our...
Yeah, it'd be good.
Maybe like to even have it on the side.
Right and then you can have like half
of your screen Notion, half of it go off bot.
Yeah, let's do that.
And we can kind of like try to burn down the tasks
as much as possible.
What do you think are the main ones?
We can help steer the agent here a little bit.
We need to...
Well, for sure, I have to get the backend shipped and deployed.
Do you want to do the client thing?
Actually, I don't know if we should.
Maybe we'll just keep it web for now
because web would be the easiest.
I guess we could maybe try like a 3D version.
Oh yeah.
For the animations maybe, just to see.
And maybe just not make that a blocking task
that can be like a little side quest
for one bot to go and just try.
And if we like it, we can take it.
If not, we'll just try.
And then I think it'd be really good
to maybe make some progress
on just thinking through the ad stuff.
I know we kind of like briefly talked about it.
I don't really know what our plans are for that yet.
Yeah, we can think about it.
So it'd be good to...
We had a cool promotion prototype.
We have a promotion for like actually running some ads.
So we have some assets.
We can think about like, okay,
if we have this like stadium theoretically
that the bots are in,
number one, what does the space look like?
And then two, like how do we,
I mean, the question would be like, okay,
how do we sell these things?
Like how do we account for that?
And then what gadgets do we choose?
Is it like a marketplace for ads?
Are we just like selling definitive slots?
Like how do we price those?
So I think that's like some exploration
to think through the framework,
but then also like we can just start to build out
that system since we have server, right?
Yeah, cool.
Okay, I got those captured as we were talking.
So I'll get those into the notion doc.
And hopefully we should be able
to pull that up on in a second here.
Let me find a micro window.
Where is my, here we go.
Working to do this.
See if I can-
Maybe ask your bot to share this doc with us
on Slack, ping us on Slack.
And actually something I haven't set up yet
that I usually do is to have a kind of like a Slack
manager bot that kind of like watches for pings.
And then it can kind of,
I'll just tell it to like,
hey, you know, every time we do a task,
update the notion.
That's great.
Maybe we can all do that.
Yeah.
Get a-
Oh, you want to try something fun?
Why don't we all turn on voice mode at the same time
and create the same bot?
I think that could be really cool.
Should we try it?
Sure, let's do it.
Let's all open up our doc.
We're gonna do the prompt.
Why don't you do it?
It's your Slack bot or your Slack idea.
Okay, okay.
All right, so on three,
we hit start, wait, wait, which bot are you on?
Are you on doc.eggbot?
Yeah.
Okay, okay, okay.
So we'll talk addressing doc.eggbot.
Yep.
Ready?
Three, two, one, go.
All right, hello, doc.eggbot.
We are interested in building a Slack bot.
So the job of the Slack bot will be to listen
for ad mentions of our username on Slack
and process them.
So there might be things like,
we might get notified of things like,
here's a task that was completed
and there's gonna be a notion doc
that's tracking all of the progress we're making on our game.
So maybe let's start with that for now
and we kind of tweak the instructions as we go.
But yeah, for now a Slack bot
that just responds to mentions on Slack.
Let's hope that works.
Let's hope that works.
Hey, it worked for me.
Wow.
The one, do you just say, restate this in your own words?
What do you typically do as
in progress to do to doing?
And then what, we'll sometimes do
Okay, yeah, I need to connect Slack
but I might do that when we're not looking at my screen
so I'm not showing any secrets.
So can we go to the wide angle shot for a second
and give me a chance to connect Slack?
Thank you.
In the meantime, I'm getting off a couple of agents.
I got some ads research,
so running like an agent to kind of understand like,
okay, like how do video games still add?
Like how do other businesses think about this problem?
And then making a recommendation
for how we'd implement that on the client
in our server.
And then also like Lauren mentioned,
checking out some like maybe prototype pixel implementations
cause that could be cool.
So I feel like pixel art sometimes is good.
Sometimes like really bad.
So we'll have to see maybe play around with that a bit.
Cool.
And I almost got Slack connected here.
I'm just doing my last GitHub setup as well.
In the meantime, if we wanna cut to my screen
I can show off some of what these
like the remotion assets ended up looking like
from the last time.
And then we could also walk through some cool
rock bot uses that have queued up.
That'd be cool.
Anything we should be thinking about
for client server implementation
while I'm not splitting that up.
Okay, cool.
So basically what we talked through before is like,
right, how we motion built these as code.
So the cool thing is that you can just like
start to create just different aspect ratios.
Obviously like, you know, is this perfect?
Like, do we wanna adjust the crop?
Like probably, but now we have like a 16 nine asset.
So you'd imagine like, or nine 16 rather,
we have one one nine 16, 16 nine.
So now we could advertise.
Like maybe there's a platform that prefers this
probably like LinkedIn, Instagram, X.
And there's like a way to easily start
like distributing these assets.
This is like perfect fidelity.
No, it doesn't really explain much about the game
or how it works.
But like once we have the assets in the game
and the game mechanics, we could then slot those in
for the react components that we have
and sort of like build out some like dynamic videos.
Right.
So the whole point here is I could pop this open
and then like render this out as a video
where we'd have like a video asset on my laptop.
So yeah, that was pretty cool.
That was really cool.
Yeah, I love remotion.
Yeah.
What are my favorite tools?
Yeah, you can just like create these so quickly.
Yeah I might be cool to even get a bot to like
put some of these in some iPhone frames
and and see how they should.
I don't know if that's like,
cause the bots are pretty good at generating just like
in like live HTML in the chat.
They are.
You can also typically what I'll do for iPhone frames
is like, if you use Figma, you can find like,
like there's an official number one,
there's there are official Apple frames.
Like there's an official Figma pack for it to do this.
And then like you can, you can like dig through
some of these things and find like good device or assets.
So one thing I think I'd have to set up separately
is like Figma, we'd have to set up like a Figma workspace.
We could pull in some of these phone frame assets.
And then actually this could be a pretty cool workflow
because bot is quite good at using Figma.
So maybe I'll, maybe I'll kick it over to that in a bit.
Nice.
But I wanted to share actually,
so we had a guest on earlier Karen,
the cool, cool Grockwite use cases sheet.
Basically had like bot print out like a morning summary
on her, so she has an internet connected printer.
And this is like personally one of my favorite use cases.
So I want to share that.
I think I want to do, cause like I tried a journal,
everybody tries to journal, most people try to journal.
Many people fail to journal.
I actually do like try to write first thing in the morning,
whether it's like creative out like about my life.
And I feel like it would be really cool to have like a template
withing date that just gets printed out every morning.
And that would like, to me that feels like almost more tactile
than opening a journal.
And I feel like it's a better reminder.
So that's something I want to play around with.
And then also I'm thinking about getting like a receipt printer
and then like hooking it up to Twitter
and like printing out tweets.
I think that would be interesting.
Another one I saw, so this is Lenny,
look, this is how I do this?
I quote,
like,
he paid a parking citation.
What was it?
I was like,
this is just like a little part
of the stuff.
This is like a little part.
So he was like,
this is just like a little part.
I think that's a good idea.
I think he did well in that period.
Like I just kind of Failed.
That's what I was thinking.
Well, I don't know why I just wanted to try to
to find out how I'm going to get it.
I don't know why I just want to try
to find out my free time on the internet.
So I sat down to get it there and I just did it.

===== END RUN 20260917-0725 =====

===== BEGIN RUN 20260917-0750 =====
mentioned checking out some like maybe prototype pixel
implementations because that could be cool.
So I feel like pixel art sometimes is good,
sometimes it's like really bad.
So we'll have to see, maybe play around with that a bit.
Cool, and I almost got Slack connected here.
I'm just doing my last GitHub setup as well.
In the meantime, if I want to cut to my screen,
I can show off some of what these,
like the remotion assets ended up looking like
from the last time.
And then we could also walk through some cool
crockpot uses that have queued up.
That'd be cool.
Anything we should be thinking about
for client server implementation
while I'm not splitting that up.
Okay, cool.
So basically what we talked about before is like, right?
How we motion built these as code.
So the cool thing is that you can just like start
to create just different aspect ratios.
Obviously, like, you know, is this perfect?
Like, do we want to adjust the crop?
Like probably, but now we have like a 16 nine asset.
So you'd imagine like, or nine, 16 rather,
we have one, one, nine, 16, 16, nine.
So now we could advertise like,
maybe there's a platform that prefers this,
probably like LinkedIn, Instagram, X.
And there's like a way to easily start
like distributing these assets.
Is this like perfect fidelity?
No, it doesn't really explain much about the game
or how it works.
But like once we have the assets in the game
and the game mechanics,
we could then slot those in for the react components
that we have and sort of like build out
some like dynamic videos, right?
So the whole point here is I could pop this open
and then like render this out as a video
where we'd have like a video asset on my laptop.
So, yeah, that was pretty cool.
That was really cool.
Yeah, I love your motion.
Yeah.
One of my favorite tools.
Yeah, you can just like create these so quickly.
Yeah, it might be cool to even get a bot
to like put some of these in some iPhone frames
and see how they showed up.
I don't know if that's like,
cause the bots are pretty good at generating
just like in like live HTML in the chat.
They are.
You can also typically what I'll do for iPhone frames
is like, if you use Figma,
you can find like, like there's an official,
number one, there are official Apple frames.
Like there's an official Figma pack for it to do this.
And then like you can,
you can like dig through some of these things
and find like good device or assets.
So one thing I think I'd have to set up separately
is like Figma, we'd have to set up like a Figma workspace.
We could pull in some of these phone frame assets.
And then actually this could be a pretty cool workflowureau
because bot is quite good at using Figma.
So maybe I'll kick it over to that in a bit
Nice, but.
But I wanted to share actually,
so we had a guest on earlier Karen,
some cool, cool Glock by use cases.
She basically had like bot print out
like a morning summary on her printer.
So she has an internet connected printer.
And this is like personally one of my favorite use cases.
So I want to share that.
I think I want to do,
cause I get to journal,
everybody tries to journal,
most people try to journal,
many people fail to journal.
I actually do like try to write first thing in the morning
whether it's like creative or about like about my life.
And I feel like it would be really cool
to have like a template with the date
that just gets printed out every morning.
And that would like,
to me that feels like almost more tactile
than opening a journal.
And I feel like it's a better reminder.
So that's something I want to play around with.
And then also I'm thinking about
getting like a receipt printer
and then like hooking it up to Twitter
and like printing out tweets or something.
I think that would be sick.
Another one I saw,
so this is Lenny,
shout out Lenny Richinsky.
Super, super big Glock by power user, awesome dude.
He paid a parking citation.
We all get parking citations,
especially in the Bay Area, it's like notorious.
So using our like payments link integration
he paid his parking citation.
So I thought this was pretty cool.
That's cool.
We might have some striped friends on the show
at some point.
That would be fun.
I got a third one here.
The similar to what Matthew Berman talked about,
this is Alexandra here
using the Glock bot to sell items on,
on like online marketplaces, which is pretty sick.
I think the hardest part for like,
I sell a lot of stuff on like Facebook Marketplace.
The hardest part about buying and selling stuff
is like you have to take the photos
and then do all the meta tagging
and like all this stuff.
It's so annoying.
So this is pretty cool use case.
And then the last one I have,
similar printing use case,
just printing out like,
if you, I mean, when you're a kid,
like everybody has seen sheets like these
and has probably has like PTSD from like doing math.
Oh homework.
Yeah.
So they just print out like arithmetic worksheets
for the kid,
which is pretty cool.
And I like that as well.
So yeah, those are some four,
four things that I had queued up
that I'd saw on the timeline
that I thought were pretty cool.
Yeah, that's so cool.
I think yeah, like one really underrated part of Glock bot
is are the routines, right?
Like that, like this, like these daily routines
are really nice.
Yeah.
But then there's also routines
that subscribe to events.
So like Slack messages is the one we were just talking
about, but it can also subscribe to things
for like, if you're doing coding
and doing technical things,
you can subscribe to like data dog,
sentry, whatever,
and have a, you know you start to think
about like a factory type set up, right?
where you have an alert.
Oh, our back end is down.
like then spin off some bots to go investigate triage.
100%.
While you're working on that,
I got our notion doc cleaned up a little bit.
So if we can pull up a Russian screen,
we'll take it, take a look at the notion doc.
So I bought starting to merge some of the back and forth.
Awesome.
I'm getting all these notifications.
So make my engineers getting all these.
I need to remind it to actually check off the notion.
Oh yeah.
Well, I was actually about to send a Slack message
to both of you.
Oh, and so here we go.
Yeah, my screen's back up.
So I retooled our notion doc
to look like a Kanban board.
You inspired me just now for question.
So here's the cupcake board.
We have like not started up next.
And I just asked my chief of staff here
to send you all a message.
And I actually really love this feature.
It just sort of like gives you a message option
and then you can just send it.
And there you go.
And so that's going to send.
And it lets us preview the draft of the message
and be able to send it.
So we have a message off to Slack for both of you.
And then going back to bake over here
is our founding engineer, seeing all those PRs come in.
Hopefully your Slack bots get that message
and then can know the canonical source of truth
for our Kanban board.
Sweet.
Cool.
Yeah, I decided I'm going to make a personal PM bot
as well to, I like to have like specialists for my bots.
So this bots only responsibility
will be to update the notion.
Great.
Then I have my Slack bot that's only responsibility
is to handle my ad mentions on Slack.
And then it's going to work with the rest of my team
to, I guess I don't know, like when I merge a PR,
update the notion and things like that.
That's awesome.
Cool.
Okay, so we have a task list going.
It looked like a couple of these are in progress.
Hopefully all of our bots are able to pick up tickets
almost like we've made our own little ticketing system
in some ways.
Can assign our bots to various tasks.
So I think that seems really good.
Seeing a bunch of PRs coming in.
How are we feeling about the actual app?
Do we feel like it's in a good spot to take a look?
I don't think it's finished yet.
Cool.
A lot of rebasing.
Any good software engineering project?
Let's see if there's any of these other things
that can maybe, we've got to wait on the play test,
working on the backend, working on auth,
the 3D client experiment.
Do you think I should send some agents off to work on that?
I think we need the server in for that first, right?
We can still do some prototypes.
Maybe you can just tell your bot, hey.
I'm going to start a new Docker agbot,
or talk to Docker agbot and make a new bot.
Yeah, you could make a prototype or bot.
I think it'd be interesting.
Maybe the 3D only needs to come in at the battle part.
Just to make it more interesting maybe.
Although I guess then the question is also
how do we translate the bot to 3D?
Yeah.
Well, let me see what our agents come up with.
Maybe I can try some things.
Because the bots are all in a standard shape anyway.
I guess the complexity of 3D is we'll have to think
about how to visually disambiguate them,
but maybe it's not a big deal.
It'd be good to just see a thorough way of prototype.
Hey, Dr. Agbot, we want to try
some more front-end client explorations
for the client application that we've built
in the cupcake repo.
Can you spin up a bot whose job it's going to be
to try and prototype ways that we can add
more 3D animations while keeping the app super lightweight?
The bot should be able to use cursor cloud agents
to get this work done and interact with the repo
that we just talked about.
I'm really interested here in being able to swarm agents
using potato mode on a bunch of different
front-end prototyping tasks.
So the core purpose of the spot
is going to be to prototype stuff.
And then as we find stuff that we like
or find examples of these prototypes that we like,
we'll try and merge those in if they make sense.
So let's go ahead and get that spot.
All right, cool.
I've got a bot going.
I'm seeing more PRs coming in too, which is great.
What did you think of the game design research
that I sent over?
Oh, I haven't had a chance to look at it actually.
Maybe we could pull that up.
You had a doc for this, right?
Notion.
Yeah.
What is it called?
Hmm.
I think I messaged you on spot.
Oh, let me ask Ping.
I'll go back here to my to-do list.
So you should have it on screen.
Oh, it's in the RDMs.
Okay.
But yeah, I think our game mechanics
are a little bit too plain at the moment.
It doesn't have to block the launch,
but it'd be good to just start thinking about,
I mean, we have some ideas in that doc.
Yeah.
But maybe we can figure out which one's the best one
to fast follow after we get it launched.
I was speeding up a bot for that.
Oh, wow, I just saw some updates
still turned on the Notion board.
So our bots are definitely doing something there.
I just asked doctor,
I bought to spin up a game design bot for us
that can handle ingesting that information.
So yeah, it looks like you did DM me.
There's the Notion doc.
Thank you.
Yeah.
And here's our new bot glow 3D.
I like using these labels too.
It just helps me keep them all organized.
And I like fun names for my bots.
So it's kind of fun to be able to do that.
So glow is our 3D prototyping bot.
And it looks like it's asking me
what we'd wanna try.
Which motion lane should we buy stuff for system tours?
What do you think?
UI Chrome depth?
My match and fight forces actually,
that's what we were talking about before, right?
Yeah.
Or I can give it a few more.
This is a multi-fix, so I guess.
Yeah, I don't know.
I mean.
Let's do a match and fight.
Match and fight could be probably the best place.
Cool.
And I mean, you bought here a crit.
That's our game designer.
Hey, we got a whole team cooking.
Watching Slack, watching GitHub.
Game designing.
Actually, I don't know that we just,
that we've told anybody this.
We just launched one password integration.
Oh, cool.
Did we announce that?
Yeah.
I think we did.
I saw it on Twitter.
Oh, no, I meant on stream.
We have it on stream.
I see it on Twitter.
I have it open on Twitter,
because our X, rather.
So, whoops, a little bit of feedback.
If we could actually cover my screen real quick,
we'll talk about it.
X.com slash bot.
All right, let's actually just write at the top.
There you go.
Boom.
One password.
So maybe we should start,
I mean, this would have been super helpful to have
when we're setting all that stuff over the one password.
We spent like a day doing that.
But the team shifts fast.
Maybe they knew.
Maybe they saw it.
Maybe they saw it struggling.
And they're like, all right, we got a trip for you.
Okay.
Well, hey, we have a whole another day of stream.
There you go.
All of our secrets management tomorrow will be such easy.
This is really cool.
I wanna try this out.
All right, cool.
So where are we?
You're working on some of this design stuff?
Yeah, I'm totally lost here.
Working on, so I have bots trying to work on the,
actually I don't know,
could you double check if that notion page
that you shared is in our notion space?
Cause my bot is having some trouble getting it.
Let us see.
It is.
Okay.
Do I need to share?
I mean, I think it was only private.
Oh, okay.
Let me change that.
Okay, you should have access now.
Cool.
I just asked my bot to try it again.
Yeah, it came out with some good feedback.
The fleet, this is a crazy dog.
The fleet?
The fleet pulse.
It's a rollup of durable facts
from active ship by Thursday bots.
That's really great.
I guess they just decided they'd wanna make this.
This is the, our bots are writing updates
from PRs that merge.
One of our bots is, yeah.
Cool, I have two bots spinning on the 2.5D 3D prototyping.
So hopefully we've got some cool examples there soon.
Those are cursor cloud agents that are running
cursor cloud agents page.
I think our backend is almost done.
So I'm gonna start wiring it up to the client.
Okay, great.
And so hopefully in a few minutes,
maybe after the break we'll have it playable.
We'll definitely try to play with it
in the background as well and continue iterating.
But we're getting really close
to being able to share the game with the chat
and maybe getting some real games and feedback as well.
So one thing I'm actually really interested in
is setting up kind of what I was saying,
like a feedback factory where we'll get feedback
from people playing the game who like,
this is boring or whatever.
And we'll get that and start to automatically triage,
put it into our notion,
get the whole system going and try to move really fast.
So yeah, my bots are wiring up the backend now.
So hopefully we make some progress.
So I was just looking at this MVP here for like an ads,
a lobby-sponsored billboard.
And it looks like it's proposing like an endpoint
for serving the billboard and kind of describing,
hey, there's no ads routes, so we need to add one.
We might need one that allows us to like slot those in.
And it's describing like a datum
having advertiser campaigns and creatives.
Honestly, it seems a little too complex.
I might slow this down a little bit.
But yeah, I mean, I think this isn't,
I don't think this seems to be too complicated.
It's interesting to see.
Oh, I'm actually also looking at the chat
and I guess there are some questions
around like what game we're building.
For the people who are joining late,
maybe Rushan, you wanna give a quick update?
Yeah, we'd love to.
Maybe even show what we have so far in the web prototype.
See if I can find my prototype.
I had it running somewhere.
Cut to Roshan's screen here.
Yeah, if we cut to my screen,
I can show you a little bit of what we've been building
in the background.
So, cool, yeah, so this is what we're building.
So we're building a game studio in our first game
and that's the business that we've been building
on stream for the last little while.
We have a working name for our game,
which is Cupcake, and the whole premise of the game
is that you'll be able to log in with X
and pit your bots head to head against other bots
in order to win the global leaderboard
and get to the top of the global leaderboard.
How did you get first place, Matt?
You know, I don't make the rules,
I don't make the rules around here.
What the hell?
How did you?
It's a placeholder leaderboard.
We'll match up for real.
I'm very competitive.
When there's a leaderboard, I'm like, ooh,
I'm gonna optimize.
That's what we're hoping we'll make.
We're hoping everyone's gonna be competitive
to drive a bunch of users to the game, right?
So this is our prototype that we built
with just a bunch of HTML
and we have a client app now
that we're wiring into a proper backend.
But the idea here is that you'd be able
to customize your team with bots that you're uploading.
So if you create a bot template using GrockBot,
you'll be able to add that bot in
that bot will be assigned some stats
subject to some rules that we built.
And if you're curious to learn more about some of that,
you can go and check out some of our sessions
from this morning.
Basically you have a team of bots here
and you create a lineup for any match.
So you choose a captain bot,
get a few friends that join your team,
and then be able to have your bots face off
against someone else who's logged in
and who's also in the app at the same time
and kinda earn some points and grow your rating
and get onto the leaderboard.
So that's the end-to-end mockup.
Obviously we're gonna continue adding to this,
but that's kind of where we are.
Yeah, what I like about this is like,
so right now it's obviously very simple,
but I think there's like a lot
of fairly interesting mechanics we can add to the game.
Like, you know, we have a leaderboard,
it's competitive.
We have this notion of like a captain bot.
Don't really know what that means yet,
but maybe that kind of some special,
like maybe there's some special difference for the captain.
Yeah, I think that could be
a lot of really interesting ideas.
Like maybe there's a chance as well to like,
if you win, like you can,
you get a chance to take the other person's bot
or something, I don't know.
You know, something like that where,
I feel like there's a lot of interesting ideas there.
That's actually my favorite part about this idea.
It's like there's so much,
so many places we could go with it, right?
So it's like, let's get the,
once we get the core out there,
we'll be able to kind of go from there.
So, cool.
Okay.
We also have a task list that we're working towards.
So it looks like auth work is in progress.
Remotion ads are in progress.
Working on the API in the back end here.
Let's see what my,
so I have a game critique bot here
that Lauren had had a bot that wrote up some thoughts
on our game design pattern here.
So just to kind of recap for folks,
we basically built some game mechanics here
for these bots to go head to head
and wanted to kind of validate, you know,
are these best practices for game design?
None of us are pro game designers.
Some of us are pro game players or not pro,
but you know, enjoy playing games,
which really nice is we can just go and like,
you know, get some wisdom from the world here.
So it looks like we have
some proposed changes to the mechanics.
Do you want to zoom in a little bit?
It's hard to see.
Yeah, let me do that.
There we go.
Drop bot full screen here for a second
so we can see that a little bit better.
Type only.
Okay, so maybe some suggestions here.
Actually, one thing I'm noticing is,
like, well, a meta thing I'm noticing is like,
some of the writing here is kind of,
it kind of looks like AI writing.
So I wonder if you could use the handy dandy unslop
skill or the bro skill in PSAC
to, let's get like a plain English.
Unslop bro.
Plain English.
These two are the most,
I hear they're the most useful skills in PSAC.
So very, very handy for,
you know, like, what did my bot just say?
Very, very useful.
Yeah, and if you have not installed PSAC,
go into the marketplace.
You can install PSAC.
You get all of these skills.
There are over 40 of them here
that will help you be a potato engineer.
Okay, so it's a clean version here.
So it's a truth, looks good, cool, ask.
I'm gonna have this crit agent
actually talk to our founding engineer
and just make sure it follows these rules.
It might be good to have a canonical game design
document as well.
So it's like one thing that's really useful
when you're starting to work in a team with Gropot
is like sharing context with each other.
And yeah, I think it'd just be good to,
like I have that one scratch doc for the game design,
but it'd be really nice to have a single doc in
that, you know, we can all reference
or the bots can reference.
Great idea.
I'm gonna set that up right now.
Send it to Lauren on that, on Slack.
What I like about this is I know
you both have bots listening to Slack.
So I can just say send something to Slack
and it just hopefully will make its way
into your bot ecosystem.
I haven't seen anything in Slack.
I'm gonna ask my bot as well.
But my bot is definitely listening.
Yeah, let me pull up mine.
So I have a bot here called Ping
that is monitoring Slack.
And if I look at Ping's routines,
yep, anytime Slack is mentioned,
when mentioned, at mentioned anywhere in Slack.
So anytime I'm not mentioned,
it should be working.
Make sure you watch DMs too,
not just at mentions in random channels.
Yeah, one thing I like to do as well
is when I start to feel like
there's almost like a workflow that I want.
That's the point where I start thinking about,
okay, how do I get Dr. Eggbot
to actually fix up not just one agent,
but all my agents.
So, you know, example,
you know, like let's just say
that we want it to be very communicative.
And like, you know, every decision we make
should be what should go on to Slack.
And, you know, we want to always keep each other informed.
That's something Dr. Eggbot can actually help you with.
So you could tell it like,
hey, Dr. Eggbot, we want to be very,
you know, fluid with our comms.
And every time we make a decision post a Slack
and hopefully that it doesn't overdo it
and literally Ping would like spam this.
But like the point being, you know,
when you're starting to feel like
you are almost like micromanaging your bots too much
with individual commands and prompts,
it's a good time to kind of like take a step back
and think like, hmm, how do I actually, you know,
think about workers trading my team of bots
and Dr. Eggbot's like really good for that.
Cool.
And it looks like my Slack bot is updated
to be listening to DMs as well.
So let's say we should be good.
How do you guys see our bots updating the notion?
Let's take a look.
So I see a bunch of tasks here.
I did see a few things flip over from up next to in progress.
Like, clerk was not in progress a few minutes ago.
And so I think we have our bots updating notion,
which is great.
I'm just gonna see here,
switch back to, okay, we have a feature here
for the replay tape, it just got murdered.
Yeah, I'm gonna tell my Dr. Eggbot to tell all my team
that the notion is gonna be the source of truth
and every bot should keep it updated.
I'll do the same.
Cool, it looks like we have my founding engineer earlier.
I asked, I had my game design bot here,
ask my founding engineer to take a look
at the client implementation and make sure
we have the latest feedback from our game design overview.
And so it looks like we were messaging back
and forth from the bots and looks like
we maybe need to fix some stuff here.
So I'm just go ahead and have our engineer just start working
doing what it thinks is best.
Okay, what are you all updating some of the ad stuff?
So we're working on that like maybe like a little bit of a
marketplace.
Also just made sure that my second missions are live
and listening, so we should be good there.
It's going, I'm not gonna lie, I'm a web client,
so that we can actually start playing the game.
Nice.
See if I can get some of these videos working.
So I had my, while we were talking, we kicked.
I kicked off a 3D prototyping, and bot thatเข,
that, that was gonna take a look at some,
as we could add some 3D to this experience
and explore some mockups for that.
Let's see if this video works here.
Maybe it won't.
I
figure out what's going there, but it's cool about this is the bot
seems to have generated some videos in the back end that try to
put some mockups together, which is pretty exciting.
Which is pretty exciting.
Agents in general are not super good at writing tests.
So one thing we haven't done yet is, I care a lot about code quality,
but at such an early stage of the project,
it's more important to get it out the It's more important to get it out the door,
but I think once we have it up and we're starting to to iterate,
I will set up the code quality, start,
start thinking about performance, things.
So tests are, yeah, I almost wonder if tests are slowing us down.
Should I delete all the tests?
There are tests right now.
Let me tell my bot to delete all the tests.
We're pushing to main and deleting all
the tests. True, scrappy mode.
We'll write the test later once we know what-
Once we know what the testing
.
I think my bots are updating the notion.
I can show off project agents.
So if you haven't cursor projects before the new feature that we launched,
either a few
few days ago or last week, I'm blanking on when exactly.
But very recently, and projects are like one of the best features in cursor.
The way I think about a project is really it's a long running coordinator agent that
does a really good job of keeping track of subagents and smaller agents that it
orchestrates to do tasks.
And so it's kind of similar to the multitasking mode that Roshan shared earlier.
But this one's like a little bit, I think projects are a little bit more powerful,
because projects have access to certain things like a share.
So one thing that projects have is that they share this like almost memory between the
colleagues.

===== END RUN 20260917-0750 =====

===== BEGIN RUN 20260917-0822 =====
agentic loop. So it does a lot of things like it will obviously have agents that go off and
implement those tasks. And then it will also spawn like verification agents that will actually
try to run the code and make sure it's correct. And depending on like how much, I guess,
how much reasoning you apply in setup PSAC, so setup PSAC is a skill in PSAC that
lets you adjust how much multitasking you do on the max level, you know, will spawn like a swarm
of agents to actually run the application and fuzz the application. And fuzz meaning like they'll
actually run it and like kind of, you know, click around, make sure that things are working.
And you have like lots of bots like almost like simulating like real users.
And I'm, for some reason, the project agent has picked up on some potato nomenclature or terminology
here. Like instead of saying merge, it says mashed. I don't know why I'm saying that, but that's fun.
It's that personality. I think, I think Tater must have told it like, yeah, mash it in when
I'm not sure. We do have a, we do have a bot here called mash. So I wonder if it's doing
that. That's great. And what's really cool too is like, while you're doing all this, I'm seeing the
notion can bend word just like all these updates. So we had a bunch of stuff that was next up that
is now all in progress. And a few things have moved over to done like they were we need to rebase
PR 21 and land PR 19. So it looks like both of those are so that's really cool. The notion
thing kind of just fly. It was kind of cool. It's like, you know, if we had an extra
monitor in here somewhere, could you like put the dashboard up on a monitor and just
slow it down in real time and that'd be really cool. So just the real ticket, the ticket burned
out. ETA to complete. Cool. We're going to see if we've made any progress on our 3D design
prototypes, kind of see what we came up with. Shall we take a look? Yeah, let me give me a
second. Let me pull up these PRs. I think I have one here. So I can probably pull them both
up right quick. Oh, I forgot to mention about what while you're while you're working on
that, that projects is really good at is the fact that every single agent on the project is
actually a cloud agent. So I could actually just go off, turn my laptop off, go and have dinner,
go get some Boba, and it's going to keep working. And oh, it looks like we have your
screen up now. Yeah, you're minding me on the Boba thing. Let me forget. Oh, yeah,
we should get some snacks. Yeah, yeah. So yeah, what I asked earlier, an agent to spin up
some 3D demos to get a sense of, you know, how we might, you know, bring in some 2D or 3D. Oh,
yeah, let me zoom in. It's not going to... Well, let me just see if I can just pinch to zoom. Yeah,
there we go. Yeah, it's a video. Yeah, I'll slow it down a little bit. And so what's cool about
this is a cursor agent, you know, prototypes and stuff. And then also went ahead and took
a video of how this would look. And so it did okay at that kind of scene. Wait, is this
supposed to be 3D? I think it's told itself it's a 2.5D tape clash. And so I think it gave itself
a couple of different prototypes. This is SLAM 3D. Let's see what SLAM 3D is. That's kind of cool.
Okay, let's see if we can... Why is it so fast? Oh, did you put it on? Well, I think it's just
it doesn't have a good sense of the game engine yet, but we'll see it just go like...
Yeah, we'll need to fix that. Mash, mash, mash. Yeah. Okay. There's another one here. That
was a 2.5D. There's another one I think we had, going... I almost wonder if you actually need to
explicitly say let use 3JS, maybe? Yeah. What's a 3D? I can, I'll try prompting for that. So
hoping we'd have some other videos to see. Okay, you can see a sense of how these would look.
These are more static, but... A lot of PRs are getting merged. Nice. We have our
BOT factory in full swing. Love it. Love it. Okay. How's our task list look and notion?
We'll pull that back up. Oh, cool. I did actually get the videos in GROCKBOT. That's what I was hoping
would happen. So there we go. We got these videos in GROCKBOT to take a look, what we just saw,
so we can actually just do this in GROCKBOT. Again, the animation speed is not perfect,
but we'll give it some feedback here. Someone in chat also mentioned that maybe we should
have some game music. I actually forgot entirely about music or audio. That could be fun to add later.
Let's actually add that in. Where's our PM? Let's add a task to the notion list for sound design.
I'm actually curious. Do we have plugins that help you make sound? How did that work?
I'm not sure. We'll probably find out. I wonder if we could just engineer some quick sound
effects. Probably do some 8-bit type stuff for you. That seems like it would be...
Would it be annoying? It would probably be annoying, yeah.
I mean, we could make it good, though. Yeah, I'm going to make a new bot that will help us explore
some sound design for the game. We have a task on the notion board, so we should be able to
just pick up that task and run with it. I want some really high-level exploration. Maybe we
could explore some styles that are 8-bit music or something that we're generating. It should be
not too annoying. I find annoying sounds are always the worst when you're especially on a
web browser and you don't expect the music to be playing. Let's spin up a bot whose job it is to
be our audio engineer and go figure out some prototypes for that. Would it be worth...
Since we're going to generate these songs... I didn't read the whole problem, but maybe
we could put them all into notion until our backlog and then we can review it later as a team.
Since we're trying to move more into tracking everything for notion.
I also like give your bots feedback. As soon as Dr. Eggbot spins up a bot here,
I'm going to let it know that we want our audio designs to live in notion.
Well, a lot of PRs got merged. What a good name for a bot.
Okay. We have tones, our audio sound engineer.
Make drop the audio in notion, BTW, so we can all listen and check it out.
Roshan, do you want to share the Eggbot promotion again before we cut back to the main stage?
Oh, yeah. I didn't realize we were losing, got so into the build
that we are going to be cutting back to the main stage in just a minute.
Let me cut over here to our Eggbot promotion. For those of you who are just joining or if you
missed this earlier, we are giving away some free Grockbot if you want to build the way we've been
building or want to be able to try Grockbot out for any of these cases you've heard today.
We showed a bunch of really fun personal use cases, being able to print newsletters.
If we can actually get our camera off of the overlay, it will make the QR code easy for
folks to see and stop covering it up. Thank you very much. If you want some free Grockbot,
follow these steps. $200 in value, so a really great deal here. The first 1,000 users,
exclusively on stream, the first 1,000 users who duplicate and create a bot with Dr. Eggbot
will get a chance to get a free month of Grockbot. Hopefully you guys get to try this out,
and hopefully you've seen us use Dr. Eggbot over the course of the day to create a bunch
of bots of our own. It's really great. I highly recommend it. Up next with that,
we're going to go to Grockbot for customer support on the main stage and we'll be
back right after that. Hey everyone. Thank you for joining today's workshop on Grockbot
for customer support. My name is David, and I am a software engineer here at SpaceX AI
in the user ops org. I'm really excited to share what we've been working on and how
we're using Grockbot for customer support. Here's the agenda for today. I'm going to
quickly give some context on Grockbot in case you haven't been watching the live stream
or been to one of the other sessions. Then I'll share a few specific use cases for it
around customer support. Then I'll demo a lot of these. My goal is that the session is a little
bit more demo heavy, so you can really see it in action. And finally, we'll finish with some Q&A.
So let's get started. So first, what is Grockbot? Grockbot is a new product that lets you
hand off real work to an AI coworker. The interface looks very familiar, just like a
messenger app on your phone or a computer, but under the hood it's built to keep working
after you've sent your original message. So you aren't just chatting back and forth with quick,
simple one-offs here and there, you're really able to delegate a real project
to Grockbot and it can go off and take it from there. One of my favorite things about Grockbot
is how flexible it is to the way you work. So you can start with a single, general bot to
help you get set up quickly and easily, or create a team of specialists that coordinate
around the task and work in parallel on certain subtasks, and however your team
operates, Grockbot can fit in and be very useful very quickly. So why Grockbot? Why does it feel
like interacting with a capable coworker rather than just another AI tool? For me, there are three
big reasons. First, Grockbot is always on. So by now you may have heard that Grockbots have
their own computer. All that really means is that they can work without using yours,
which is a little bit more powerful than you would think. So if you close your laptop or you
are working on something else, your Grockbots can still finish the task you assigned them without
you having to check in constantly and steer them too much or slowing down your local machine.
There are also routines which are cron jobs or event-based triggers that you can launch
your Grockbots with and make them super reactive. This way you don't have to be online at the exact
moment you want them to run or start working on the task. These are super simple to set up. All
you have to do is tell it, create a routine at this time or on this event, just like a normal
conversation and it will set it up for you. Second, it is really easy to use. The interface is
familiar for a reason. You don't have to be an engineer that is familiar with the ins and outs of
an IDE to use bots effectively. You can just talk with it normally like you would a friend or a
teammate on Slack. And finally, it fits how you actually work. There are already connectors
for most tools you use. You just have to go in the marketplace and search for them.
There is likely one for you. You can plug Grockbot immediately into your ticketing system,
whether that is plain, Zendesk, Intercom, where your team chats like Slack,
where your knowledge base is like Notion. From there, you can use the bots and routines
like building blocks that fit the way your team actually works. And one really powerful thing
about Grockbot is if there is a missing piece, you can likely have Grockbot build it.
Grockbot has access to cloud agents. So if there is something that you are missing, you can
just tell it to spend up a cloud agent to build a certain connector or a certain thing
and it can go off and do it for you. So you can get creative and build around and you are
not stuck waiting on Enge roadmap or a third party vendor to add support for the thing you need.
Cool. Now we will get into some customer support use cases.
So the big one is always candidate answer support tickets. The answer is yes. And I will show
this later in the demo. I know this can be scary at first to let an agent
sort of reply to your customers. So you can definitely build up to it.
You can crawl, then walk, then run. At first, you can have it just read tickets,
maybe give you a summary and the root issue of what the user is asking about.
And maybe it can draft a response. Then you can have it add the draft to a note in your
ticketing system on the ticket so you can sort of see it before you press send.
And finally, when you are confident, then you can have an answer and actually
respond to the ticket and the user. There are also things you can do around the guard
rails and environment you give GROC bought. So you can set up evals and traces fairly easily
so you know what it is thinking and what actions it will take
before it publishes. So when you can add traces to those draft steps and then
sort of make sure it is behaving the way you want it to before going on.
Cool. The second one. So at scale, you may have a lot of support tickets with a ton of
different customer states and it can be really challenging to find the things you should care
about in that exact moment that needs your attention. GROC bought can make this a lot easier
through alerting. So you can spin up a bot that looks for certain things.
For example, maybe your team is hyper focused on churn this quarter and you want to know
when someone writes in that is threatening to churn that has been a customer for six months
or more. Anyone on your team can spin up a GROC bought that you have connected to your
ticketing system. You can just tell it to create a routine that looks through your tickets
every hour. It classifies if a user is threatening to churn and has been a customer for a certain
period of time and then send you an alert in a Slack channel so everyone has visibility.
Another one that maybe is more powerful than you think or is harder to get to than you think
is an internal question and answer agent. So if you have used other support agents,
they are primarily focused on customer replies. It is surprisingly cumbersome to leverage
the knowledge base you have already created or accumulated to answer your own teammates'
questions. You may have someone in GTM that is trying to get up to speed on today's release
about cloud agents so they can come prepared to a customer call or a TSE manager looking for
changes to the internal refund policy over time and if that has impacted certain customers.
You have already done the hard work of creating this knowledge base of both public info, so
thank your docs and help center, as well as likely some internal info. So you might have
suggested operating procedures around how your teammates handle refunds that is not public
information. You should be able to query this intelligently within Slack and GROC bot makes
this really easy. And maybe the last one and one of the cooler ones is it can improve itself.
So finally you can give GROC bot your traces or have it look back on the last week of tickets
and sort of ask it to look and try to find improvements for tickets that maybe could have
been handled better, could have been flagged earlier and it can go and do those things for you.
Cool. So next we're going to get into the demo. I have created this demo. I have a team of four
bots. They are build so it sort of runs the setup and the infrastructure reply. So this
actually answers users in both our ticketing system and then Slack. There is an alert. So this one
usually gets pinged by reply and posts an alert and Slack and tags me and then tune which
tries to improve the system as we go. I want to call out that you definitely do not have to
have these in mind or set them up before getting started. In fact, I don't recommend that.
Instead, you can start with one bot, name it a generic name and try to teach it one workflow.
Then when the scope grows or you need more than one running at a time,
I recommend breaking them up naturally then and reorganizing them in a way that makes sense to
you. Cool. So now I'll get into the demo. I have GROC bot set up with the bots we mentioned
and then I have a few things that might look familiar if you are inside of a support org.
So first, we have a knowledge base. Right now, I just set this up in notion as an example. So we
have public docs. So think of this as like this could be your documentation or your help
center that has a few things on product and common issues like auth and billing and an FAQ.
Oh, I should have mentioned that I'm going to use the fly low example that you may have seen
in other presentations. So this is basically like a pretend airline and I'm going to simplify it.
So the only skew we have is a monthly internet subscription that is built monthly and is $20 a
month. Cool. So we have this public information that a user could find if they search Google.
We have some internal policies. So right now, this is stuff that we want maybe a billing support
member to know if you have a human answering, but you don't want to share broadly with the public.
So this is just a very simple example of if a user asks for a refund, you can approve it within 14
days and cancel and refund if they've been subscribed for more than 14 days, deny the refund.
And then finally, I just have a little bit of process for the Glock bot agent reply to use
when replying to some of these tickets. So we just have this loop of what it should do,
read the ticket, look for the knowledge and then decide to reply or hand off and then
act, leave a note and yeah, cool. This is plain. This is our ticketing system. So I've already
pre-filled a few example tickets and we'll walk through those. I have Stripe pulled up as well.
So I mentioned there's a subscription that users may pay for that is the basically like
the Wi-Fi pass on a plane. And then I have Slack that I'll pull up for later.
Cool. So first, let's start with maybe this first ticket. So I'm going to go through Alpha
Medical. So this is just a very basic question. I forgot my password. How can I sign in?
Or how do I reset it? So let me ask Glock bot to try to reply to this. So like I mentioned,
I have the Glock bots I mentioned. There is build. So I've already set up some of the
connectors to make this a little quicker. So I've added plain. I've added notion,
super base and Slack. If you aren't familiar, you can just click the marketplace and search
and click and install. Cool. So let's tell reply. Hi. Can you reply to Alex in plain?
And it is going to go off and try to reply. So let's see. It's going to look it up.
I pulled up. So basically what it's while I wait for it to reply,
it's going to go through this process. So I've sort of made this loop pretty simple of just
what I wanted to do, how I wanted to behave and sort of show some thinking.
And we'll see that it replied. Okay. Cool. It replied. It said replied.
If I go back into the ticket, you see that it followed the sort of tracing or it followed
the process that I gave it. So it showed some of its thinking where it says it can reply with
high confidence, the root issue and the source that it referenced to answer the question.
Cool. And then it gave the right information. So these are the steps on this public auth
forgot your password. So if we go to public auth forgot your password.
These are the exact steps. Cool. So that's obviously a pretty basic example,
but let's go on to a harder one. So this is an example of something that is not covered
in the toy knowledge base we created. So some SSO octa stuff. Let's see. I'm going to try to
prompt it so it will try to reply. But in the instructions, I told it to not reply unless
it's pretty confident. Try replying to Ben. And once again, it's going to go through the
exact same process. If I were to change something in this sort of loop, it would,
I've already told it to make sure to read this every single time. So if I were to add
maybe seven and some other step, it would know to do that after every time.
So let's wait on that for a few more seconds. And let's make sure I have
Okay. Nice. So it says it tried to do that, but there's no public docs or internal policies.
So low confidence and hand off,
which is exactly what we wanted it to do. And then you'll notice that this is new. So it
messaged alert. So I've set up that if something, if it looks like maybe an enterprise customer
that's locked out to make sure to alert me so I can know in real time. So it messaged alert,
alert activated and posted to Slack in our alerts DG channel. It is this one.
So yeah. So now I am without too much work. I can already see like in real time that
important tickets that I need to maybe action on right now. Cool. Let's go on for now. So I'm
going to mark this done for now and go on to the next one. So the next one are going,
the next one is going to be two refund related tickets. I mentioned at the beginning
that we have an SOP specifically around that. So we have two users, Carter and Damon.
If I go into Stripe and click into them, I've set it up where Carter just subscribed
and is requesting a refund. So today is September 16th. So his renews in exactly a month. So we
should grant him this refund. And then Damon, you'll see that it renews in about 10 days. So
it's already been, call it 20 days, which is, and our policy is only 14 days within. So
it should deny Damon and grant the refund to Carter and it will, we've already set up Stripe
so it can take these actions for you. You can set it up to have you approve the actions or you can
tell it, let it like run loose. Okay. So let's say, okay, now try to answer Carter and Damon.
And this one may take a second because it's going to run through the loop,
sort of make the call and then if it needs to, it will have to run the actions through Stripe.
Let me have those pulled up and we'll see this state get changed here shortly. Let me pass
in the customer IDs. Maybe I'll make it quicker. Let's, it should be on its way now.
Cool. So it's thinking, it found them now and it says both replies already up. So let's go look
at them. Cool. So it just answered for Carter. Like I mentioned, it should have canceled an issue
to refund. It's saying that it did that. If we go to Carter in Stripe and refresh,
the status should have just changed. So nice. It went from active to canceled and this payment
has been refunded. And then that is what we, that's what it said it did. And then
for Damon, it says we have denied your refund. It didn't give like the full SOP. So it didn't leak
the internal information. It was just vague. And then it says, I can cancel, I can schedule a
cancellation for you at the end of your period. So you don't get billed for the next month.
And then if we go to Damon and refresh, this should all stay the same. And it does.
Cool. Great. Okay. So now we'll go on to the next one.
So Elena is asking if we can, if she can share her wifi pass with another person.
This is something that's not actively in the knowledge base. So let's see how it handles it.
It would be under public docs. Let me move that. And it would likely go here. Cool. So let's tell
something you want just to make sure you have the final say over what happens
happens because if this were to go out without you looking and you didn't check it it wasn't

===== END RUN 20260917-0822 =====

===== BEGIN RUN 20260917-0850 =====
refresh, this should all stay the same.
And it does, cool, great.
Okay, so now we'll go on to the next one.
So Elena is asking if we can,
if she can share her Wi-Fi pass with another person.
This is something that's not actively in the knowledge base.
So let's see how it handles it.
It would be under public docs.
Let me move that.
And it would likely go here, cool.
So let's tell reply, let's go that out.
Great, can you reply to Elena now?
And it should go from there.
So it's already come back with some thinking,
found the ticket and it's checking,
it's sort of already classified that past sharing
is the root issue and looking if it's in public docs.
And now it's saying that it couldn't reply yet
because public docs and FAQ doesn't have past sharing.
And it's recommending to tell Tune,
which is the self-improver to add it to the knowledge base.
So yes, we'll say add past sharing to the FAQ.
To the FAQ, say this is not allowed.
So this is an example of it asking
before it doing something by itself.
So for your knowledge base,
this is probably something you want
just to make sure you have the final say
over what happens because if this were to go out
without you looking and you didn't check it
and it wasn't correct.
And a hundred people asked about the same thing
then yeah, that'd be an issue.
Okay, so it says it added it to the knowledge base.
I told it to do it in green so we can delineate it
so and adjust it so past sharing.
No, you cannot share your wifi pass
with another person for simultaneous use.
Great, let's see, I actually had one more thing pulled up.
Okay, actually, let's skip that then.
Cool, so back to GrockBot then
it's, we've added the knowledge
so now let's try to respond.
So in the ticket, it left a note saying to hand off
that the information was missing
and it still needs a response.
So let's tell reply.
Okay, can you try again now?
That the information's added and it says it's answering
and it says it's replied now.
So now we have a new sort of thinking note
now it can reply with high confidence.
There's a root issue and it links to the new section
and docs that covers it.
So you cannot share your wifi pass with another person.
Great, that's exactly what we wanted.
Let's mark this one as done.
Cool, that covers a lot of it.
Now let's go to one of the other use cases I mentioned.
So if we go into Slack,
we've already seen that it can send us alerts
but we have another use case where
you can have your teammates ask the bot
you've already set up to answer customers
and the knowledge base to answer like your own teammates.
So let's say I am someone on your billing team
and I want to know what is the refund SOP?
Refund SOP, so this is, I'm an internal user.
I'm asking a question and it should give me
the internal knowledge for it.
So that's the most helpful.
So we've sent it, this may take about a minute,
but yeah, we should see us getting a response here
pretty soon.
So it's asking for permission to post in Slack.
Let's say, always allow for this
and it should reply here in a second.
Let's try that again.
Make sure it's running.
Did you see the message?
Okay, it's going.
I ran this right before so you can see
that this is what would happen.
It would reply and go from there.
Let's give it maybe 10 more seconds
if not we can go on.
Now it says it's replied.
Great, so it added the emoji and it replied.
Awesome, so that's one of the other use cases.
Cool, so that wraps, again, cut the demo there.
So what we learned, first you can
and should feel comfortable setting guard rails
for your bots.
You see, saw maybe at the end there it got,
it was even more protective than I wanted it to be,
but that's probably better for writes.
You can have them be read only to start
and then work up to writes
and have manual approvals for certain actions.
You can also scope permissions to your different bots.
So one bot might have permission
to post automatically to Slack, one might not
and make them approve those actions before they do them.
Second, start simple.
So the whole product is still so new.
I don't think there's one meta or one clear playbook
and the product is meant to be very flexible.
So try to make the bots fit into how you work
and not vice versa.
Third, try to get a little creative and experiment.
You can really have it build what is missing.
Cool, thank you so much.
That is the rest of the presentation
and we can go into Q and A.
Oh yeah, so any questions?
Yes?
What's the price of like for this?
So it uses your usage that you would have
through any of your grok subscriptions
or currently cursor subscriptions.
So it just tracks usage.
I'll say it can really depend on how involved your loops are.
So if you have classifiers that run before every ticket,
if you have it write traces and evals before doing that,
that can also change.
Right now, the way I use it,
it's around $1 to $2 to answer more medium
to complex tickets.
And there are ways, I've already tried to experiment
and found ways to get that down a lot.
So there are a lot of maybe low complexity billing tickets
of like a user just asking for a refund
or a user asking about a certain new email they received.
You can sort of bucket those.
You can like run a script, have it run a script
to find those first or run that manually.
So it has more of a contained sort of set
and then have it reply all at once with the script
to those tickets.
And I found that I could get that down
to about 20 cents a ticket
for those like low complexity tickets
which is a huge sort of change.
If you've used any of the other sort of support agents,
you know that like the cost per,
they usually charge per resolution
and that's easily at minimum a dollar
if not between $1 to $10 like order of magnitude.
And if you're having humans respond to those tickets
that can be even higher, noticeably higher.
So, and that's just like with half a day
of trying to improve it.
I think there's a lot of gains and unlocks there especially.
Yeah, you're good, you're good.
Sorry.
How would you interrogate what's going on
if something goes wrong,
especially with like communication
in between multiple bots?
Communicate, yeah.
So one thing I didn't show in the demo
is I set up evals and traces for every run,
especially for the reply bot.
So this is pretty basic.
So all it is is if you have like a Postgres database,
tell it to like create a table for eval runs
or for a set of evals.
So things you want to test every single time
you like make a change.
And then another table for traces.
So every time reply runs, I ask it
how long it took, what files it looked at,
what files it ended up choosing to reply.
So usually it's like a filter down list.
And then from there you can usually decipher
where it went wrong or things like that.
And that's something you set up
in the reply bot configuration
or is that like a meta bot
that's looking at the whole system?
So I set it up just with a connector with SuperBase.
So I didn't show it in the demo
but I just set up a connector to the build bot
with SuperBase, told it has access to Postgres now,
create traces and evals table right to those tables
every time I ask you to run an eval
or create a new eval or run a trace
or answer a ticket even if it's internal, external,
even if it's a dry run ticket.
So even if it's just leaving a note on the plane ticket
it's still writing a trace
so then I can have like a log
of how it's thinking about it and what it's looking at.
And then you can also have the, once you have that
you can also have the like self-improvement bot
look at those and try to improve the system.
I think that's a big unlock.
Awesome, thank you, yeah.
Yeah, I was thinking about customer support
with like phone calls.
I'm curious if you've,
I've seen like someone has set up
GROC bot with like a phone.
I'm curious if you've seen any interesting cases
of that yet or if you've played with it at all.
I haven't done it too much.
I know Matt on the stream has set it up.
He has a good video on Twitter of like giving it,
giving his GROC bot a phone number
and I think trying to make a reservation with that.
I think that's something that's like,
definitely will be the case and want to explore.
So if you do it, let me know for sure.
I'll play around with it for sure.
Cool.
How do you guys think about implementing this
for non-technical folks in the enterprise?
Like, cause I kind of concerned if a non-technical user
is writing stuff to Postgres or without like guardrails on.
So the, sorry, I might have misheard the first part of it.
You're asking about non-technical users
and how they would plug into this.
And if there's any concerns.
Or I guess like, cause it's relatively easy
to spin up a bot to do whatever you need.
And how do you put the guardrails
so that they don't like mess up
like the production databases or anything like that.
Right, cause it's relatively easy to be like,
oh, refund all the customers that should be refunded.
But then they don't know the technicalities
of some of the backend stuff.
Yeah, that's a great question.
I know there's a lot of sort of effort
into making this more multiplayer shortly.
Right now I'd say like, whenever you first set it up
to set the guard, whoever's first setting it up
maybe can be like slightly more technical
or not even more technical
but just aware of sort of like system architectures
of like, okay, if someone is gonna run this CrockBot
I'm going to give them a template.
And someone non-technical, I'll share a template
with them that says, oh, you can't update
the knowledge base by yourself.
One thing that I've tried that I like a lot is using,
I use Notion as the knowledge base
sort of just as like an easier example
but you can use GitHub or Origin
and that's really powerful
because the primitives of GitHub are like,
you get PRs, you get code owners, you get branches,
all that stuff.
So instead of this knowledge base
where there's only sort of one
you could give someone a template on their
like self-improvement bot, for them you could say,
oh, you have to create a PR,
it has to get reviewed by Bugbot,
it has to ping me, I have to approve it.
And I can run the building blocks of like evals and traces.
I can tell it to run the evals
against a specific PR branch instead of just main
so I can have the output of that,
have some confidence that it's going to behave,
how I want it to behave
and then make sure it gets approved by the code owner
or the person that's most in charge or most technical
and you'll have confidence then of,
they can still make updates, they can still be helpful
but they're not sort of,
you're not taking on a ton of additional risk.
Awesome, thank you.
Yeah.
One question for Max, is it more talk inefficient
to queue and bulk ask requests?
Sorry, what was that?
Is it more token efficient to queue
and bulk ask requests, just bulking together?
Yeah, so it can be.
So instead of, I guess if you saw how I was talking to it
of just like telling it to reply to Alex,
that is pretty unclear.
So I could have been more clear
by like giving it a set of ticket IDs.
Whenever I tell it like that,
it has to like go look, go find, maybe read,
call the plane MCP, get all the open tickets,
look through, do a string search on Alex,
try to find that, that is a lot less token efficient
than giving it more hard details
that can look up easier.
I guess one last question for Max.
As Roshan, Matt and Lauren have been creating
their company online, do you have any recommendations
on what to set up first for a customer support team
using Rockbot?
Yeah, that's a great question.
I know maybe their sort of business has changed
a little bit, but I'd say usually there's about like,
there's usually like an 80, 20% of use cases
that create 80% of the ticket volume.
I try to identify those and have something in place
and make sure those are in the knowledge base first
and have Rockbot have really good knowledge
and SOPs there and run evals and test
and make sure everything it can handle it is
and then put Rockbot in front of it and try to answer.
Cool, amazing.
I think I am out of time
and I'm going to send it back to the studio.
You're right, Pakistan.
Yeah, I can actually share my screen.
Awesome, welcome back everyone.
Hope everyone enjoyed that last talk.
We're winding up our day pretty soon here
and I think it's been a pretty fun one.
We got a bunch of stuff done.
We have an app running and we're really excited
about some of the progress we've made today.
We thought we kind of wrap up just kind of showing you
some of the things that we were working on
in the background during that last segment.
A few things that we kicked off before we got started
if we want to bring my screen up here on the menu.
We kicked off a few bots that were working
on a number of tasks.
Again, for those who are just joining
we're building a gaming company and a gaming business
as part of our 72 hour live stream here
where we ship a product by Thursday.
And what's been really fun about this is actually
we're starting to get a little more creative with it.
And so one of the things I did before we went off
onto the last segment was spin up a bot here
that's our audio engineer
who's been working on a bunch of cool audio.
And unfortunately, can we get a mic for Rockbot?
I think that'd be really helpful
so we can kind of go through some of these.
Oh, love it.
Okay, I turn this up as loud as I can.
Thank you so much.
So our bot here has been dumping some audio
into an ocean dock.
We thought it'd be fun to go through these live
and see what we came up with.
So we have some tones that we didn't like
but let's take a look at some of these lobby songs
for our game.
Oh, quiet.
That was very calm.
It does sound really calm.
Very soothing.
But this is kind of wild.
The bot just kind of came up with these.
And if you haven't seen it before,
this is using strudel, which is really cool.
So I just told the bot to learn how to make strudel apps.
And you can just go in here and hit play
and just kind of make some pretty cool music.
Maybe not the exact vibe for our game.
Since I think we want something a little less
chiptune-y maybe or a little less pixel art.
So what I also did was asked the,
asked my bot here to give me some prompts for Suno
and just generated some tracks on Suno.
Let's see how some of these sound.
All right, that's a little too sad.
It's a kind of song for you.
Let's see what else we got.
Let's see what else we got.
I don't like it.
That feels more on for it.
That feels more on for it.
Oh, it's also a little like whimsical.
Very whimsical, yes.
I feel like I'm a wizard about to go on a quest.
OK, those are pretty similar.
All right, we're not a romcom.
Well, and then right before we came on stream,
I tried to generate a few more.
I haven't actually heard these before.
Let's see if they work.
Oh, wow.
Damn, that's some good battle music.
I kind of like that.
OK, a few more.
Oh, this is good for the draft room
while we're kind of waiting and you're building your team.
Draft room energy.
OK, I kind of like these.
I feel like we got something here
that we might be able to play with.
Yeah.
Thank you very much.
Grockbott microphone.
Thank you production.
Awesome, yeah.
So, yeah, I feel like we have some pretty good art direction
happening, right?
We have some UI that we've built.
But maybe to kind of show where we're ending up, Lauren,
you want to show how the app is looking?
Yeah, we switch to Lauren's screen on the screen.
A lot of things don't work yet,
but we've hooked up the back end.
So I've logged into my real X account
and there's a leaderboard with nobody in it yet.
I can start a new match.
I don't have any bots yet.
And I just realized there's a bug.
So I can't actually add any bots yet.
But when this works, I will be able to import any bot with a link.
I'll probably have to think a little bit about how to like.
I feel like this UI is not very user friendly,
but will iterate.
But yeah, for now, you will just like, you know,
paste like Dr. Eggbot link in here.
You'll create it and it shows up
and you can actually start playing a game.
But right now, yeah.
This is really cool.
While you're doing that, I just logged in as well on my end
and I can log in with X.
Oh, OK.
So we have logging in with X working.
Oh, did you log in already?
Yeah, I just did.
So we have atrocious underscore S.
I see.
Well, yeah, I think I need to fix the leaderboard.
OK.
I would expect the two of us to show up.
Yeah, yeah, yeah.
We're both gold.
Yeah, we're both gold at 1,000.
But hey, we both have accounts.
That's pretty cool.
We didn't have accounts a little long ago.
And very soon, maybe tomorrow, we'll come back with,
I think we're going to get this deployed.
Maybe think of an actual good name for this game.
And start playing tomorrow with chat.
Yeah, so if you're excited to tune in tomorrow,
we're going to get everyone in this game.
We've got to make it go viral.
We have one day to grow our business.
Yeah, we're going to iterate.
We're going to make it really, really fun.
And we'll get feedback.
We'll have a little factory going.
Yeah.
And we'll try to make this game really good.
Yeah, PRs are being merged.
Like our bots, we had some bots earlier
that are kind of listening for all the PRs.
And it's a constant stream.
I don't know, Lauren, if you want to show us what you're doing.
Well, I mean, my bots are just chugging on.
I have so many bots now.
Since we last talked, I have, well, we
have Wisk, our game designer.
We have Crumb, a new bot that I just made,
who is our play tester.
Who is actually just going into the application
and just clicking around, making sure it works.
So my idea here is that I want to have a play tester
bot that's just going to keep playing the game,
checking for bugs.
But I haven't taught it how to give feedback
on the game itself.
That would be really interesting to set up.
And then I also obviously have, oh, another thing
I kicked off, which I recommend doing,
since we're kind of close to the end of the stream
very soon, is I often talk to Dr. Eggbot.
As I create new bots, just think about, like, here's
what I told it.
Zoom out from a high level.
Consider my goals.
Review all our bots.
Like, where are our bottlenecks?
� how can we speed up and tighten my workflow?
And then Dr. Eggbot will actually
look at all the transcripts, the chats of all the bots
that you've created, and analyze it
and try to figure out, like, where can you
improve your workflow?
So it's got a few things like bottlenecks.
The serial factory, human merge, that's a bottleneck,.
It merged, so it waits on me.
Lauren has the interrupt bus.
So I'm still kind of playing a lot of...
I'm still very in the loop right now,
which I sort of wand at least while we're prototyping.
But as we start to get feedback,
I'm going to try to make this more autopilot.
The autopilot part is super cool.
We get to basically be directors of features.
We want to do X and then the engine behind this
I'll build features for us, make sure it's play tested,
do our QA, make sure there's no regressions.
I'm really excited when we get to that.
I'm hoping tomorrow we can just rip a bunch of features in.
I think this will be really cool.
Yeah, yeah, yeah, yeah.
Matt, any updates on your end?
Yeah, so we've been working on the ads side,
just cranking along there.
And then I also was just testing out the app.
So I'm going to have the ads marketplace up.
Hopefully tomorrow we're going to crank there.
This is my writing bot template that is currently open.
But we pop on over to, I think the ads MVP.
Haven't tested this out just yet.
We're going to need to take a look at what we're doing in the database
and probably run through some of this.
I also haven't refactored or basically
re-based on to all the changes that Lauren made.
So a little bit behind here.
But that'll be a sprint for tomorrow.
Nice.
Cool.
We had a pretty good day, right?
I think we ended up making a lot of progress.
We went app to an app, and not just an app,
but like an app with an engine and a bunch of bots.
I've basically done all the work for us today.
We got to spend some time looking at cursor,
which I think was great, and getting to use some agents
and show off things like multitask mode and design mode.
We got to make some music.
It's pretty fun.
We talked to a bunch of fun guests.
So yeah, overall, I think it's been a really great day.
And pretty excited to come back tomorrow.
Kind of a bug.
I'm fixing that, yeah.
No, you're already on it.
I want to say good job then.
Yeah, very soon, like tonight,
I think we'll have like a few games.
OK.
Maybe we can see how fun it is and kind of get
some cool ideas for tomorrow.
Love it.
Before we wrap, wanted to switch back to my screen
to remind you all that you can win some free GROCKBOT.
So if you want to build what we've been building,
build the way we've been building, install P-Stack,
install Dr. Eggbot, become acquainted with all these bots
that we've created today, you can follow these steps
to get a free month of GROCKBOT, which is $200 in value.
So a free month of GROCKBOT at a really great tier as well.
You just need to duplicate and create a bot with Dr. Eggbot
and use this QR code to do so.
You will get a free month of GROCKBOT.
This is only available to our first 1,000 users
who claim to be offered exclusively
for folks who are watching on stream.
So if you want some free GROCKBOT, get in there.
Definitely get yourself set up.
We'll be back tomorrow to share even more of how we continue
to build this app.
And I'm excited to finish wrong.
Sounds good.
Thanks, y'all, for watching.
Thanks, chat.
See you tomorrow.

===== END RUN 20260917-0850 =====
