Grok Bot 201

0:01Yeah, let's go ahead and get started. All right, let's go. Uh just first and foremost, I just wanted to introduce myself. My name is Scott Lerner. I'm on our AI deployment team here at SpaceX AI. Uh today today's session is really just to get you more advanced functionality on Grok Bot. Um I hope many of you got to attend our 101 earlier in the in the week and now today Joseph is actually going to be walking through some of the more advanced features and way to take your, you know, initial deployment of Grok Bot to sort of the advanced functions and sort of how we're using it on our end. So,

0:30without further ado, I'll let Joseph introduce himself and jump into things. Sounds good. Hey guys. My name is Joseph. I'm on SpaceX AI field engineering team. Uh so, I'm very excited to show you guys Grok Bot 201. Uh before we get started, we have some housekeeping. Uh so, if anyone have any questions, feel free to drop in the

0:48chat. We'll love to answer them. Uh before we get started, I would like to actually walk through the UI with you guys first. So, I'm going to go ahead and switch tabs over here and head straight to my Grok Bot. Uh let me go ahead and open that up. You guys can see this is a familiar uh UI that you guys been playing with around with for the last couple of weeks. Uh on the left-hand side, we have lots of bots that I created for this demo. Uh you see a star right here. It's a primary bot. This primary bot basically has all the context, all the

1:21information about all the other bots that are doing. So, uh as we can see here, we are live today. So, I can actually say something like go ahead and post on X. Tell the X bot to say that we are live with Grok Bot 201. Go ahead and send that in. So, what's going to happen is my Roy, which is my primary bot, is going to talk to X and X has my X logged in and I will go ahead

1:52and post it on X. Uh I'm going to go ahead and say uh earlier I did a test, so it's saying that, "Hey, do you want to keep the existing post? I'm actually going to say uh post a new one. Post a new one. Go ahead and send that in. So, while the bot is working with it, I can walk you guys through that with Grok Bot, it's extremely simple. The interface is just like it's just like iMessage. You'll be able

2:23to just chat with each one of the bots. Each one of them has their own memory and then it has their own computer as well and you can open these up. So, let's actually look at X over here. You can actually see and open that up and see that this particular bot, this X bot, is actually going to post that post

2:44I told it to do. This is my X account. It's signed in and it's just going to control the computer while we talk. So, if you guys so you guys can see that the bot is controlling it. I'm not touching it at all. So, while that's doing that, let's jump back into our presentation, right? Let's go back to our presentation mode and keep talking through the slides. This is kind of our agenda right now. We're going to first talk about we're going to welcome you guys, which is what we're doing now, and then we're going to talk about our Grok Bot 201 demo, which is sticker shop, kind of

3:19like a foreshadowing. We're going to go through some demos of how I created that sticker shop and then we're going to turn that sticker shop into a Grok Bot sticker shop. Lastly, we're going to end with Q&A. Does that sound good? Sounds great and yeah, while you're at it, go ahead and look up Joseph on X and go ahead and give him a follow. I just want to promote yeah, Joseph's brand well since he's he's being so nice to do

3:42this session with us today. Yeah, I do have to say I I haven't been extremely active on X. So, if you guys see less posts, just just so you guys know, I'll ramp it up. [laughter] So, what we'll learn today, this is probably what everyone is interested about. We're going to learn about how to make bots run on their own. You don't have to prompt it at all. You can have routines, you can have skills encoded, multiple skills actually, and it will just pick the right skill because it's smart enough to do that. Number two,

4:14have bots working as your teammates. Uh it's it's it's like a team and then you can actually share that teammate with other people. Extremely powerful stuff. Uh the third thing is learning that bots actually have their own memory. Uh they have their own memory. You talk to one specific bot that gets assigned with a persona. That persona has its own skill, has its own workflow, has its own

4:39routine that will be able to be invoked. So, let's do a recap before we jump back into the 201. Uh the recap of uh Grok Bot 101 was uh performed by our uh field engineering team, Martin. He is extremely incredible. Uh what he did was he created uh he printed out stickers directly for Grok Bots. So, what he did was he had

5:04three Grok Bots. One bot is called Dex. It's a designer. It pretty much grabs all the assets from Figma and then turn it into uh more creative assets using uh Grok API. Uh number two is Maya that has a that manages the community aspects that generated QR code, that did a live uh demo and showing people what it looks

5:27like. And lastly, we have Penny. Penny is uh a sourcing that found vendors that helped us print out the stickers. So, we came up with these very beautiful stickers. You know, I know which one do you like the most? Ooh. I'm going to go with the blue one. The blue one? Yeah, the blue one was

5:46Cloud app. My favorite is actually the stack of three. Three. That's that's my favorite and I think we did a online voting and then look at you was was number one. Uh definitely got uh got got tampered by Cliff there. That's that's my opinion. [laughter] Uh the the the the public knows best. So, um we have these stickers now,

6:06right? We printed out. But, what's next? What what are we going to do next? How can we build on top of this to make this even a better product? Um we can actually go ahead and so in 201, we can go ahead and create our sticker shop. Uh this is a platform where people can go ahead and uh potentially order the the stickers and create more stickers by

6:32using Dex, which is what Martin created. Um with that being said, I think we should uh go ahead and meet our uh teammates. Uh so, in our teammates of Grok Bot 201, we again reuse Dex uh to help us uh to go ahead and create stickers. Let me go ahead and jump back into Grok Bot right

6:55now. Uh let me share go ahead and jump into Grok Bot right here. You can see that Dex right here, right in this icon, it says by Martin. That's because Martin actually shared this particular bot to our team. It became a team bot. It became part of our team that anyone is able to uh grab and talk to. And how you can do that is you can do a plus sign right here, team bots. If you have your team, uh let me go ahead and search for Dex. You can see it's Dex by Martin. Takes me directly to here. You can see right before a demo, I

7:33actually uh I created a couple more. Uh so, I'm actually going to create one right now. What's your favorite color, Scott? Oh, mine is blue. Blue? Yeah. Yeah, let's do blue. Let's do a very happy blue one. I like it. Yeah, that sounds good. Yeah, okay. Let's go ahead and send that in. So, this is actually a bot that was created by Martin and he shared it to me. I didn't do any configuration to it. I just take it and then it's part of my team now and I'm going to go ahead and say, "Hey, let's let's create a happy blue one." And what's what's even great

8:05about this is Dex could be part of your team, too. What I can do here, I'm able to go ahead and create a new group chat, and then I'm just going to put decks in there. And and you know who else I'm going to put? I'm actually going to put Martin in there as well. So, in this chat right here, I can actually see that I have a

8:26group chat between the bot Martin Sorry. The bot decks and the human Martin. I totally totally fumble that one there. It It starts to get confusing. That's the whole point. We're getting our our bots to become do human jobs. It does get a little bit confusing sometimes, but let's see here. Let Let me go ahead and grab the happy blue one right and I'm just going to say, "Hey, um decks, let's uh go ahead and create a happy blue one and then at Martin and be like, Yo, what

8:58do you think about it?" Go ahead and send that in. So, decks is going to on the fly generate a happy blue one and at the same time, you know, it's kind of like iMessage and kind of like Slack. So, I'm just going to at Martin. If he's online, he's going to respond to me and I'm just going to say,

9:15"Hey, how how are things going here?" So, uh this might take a while. It might take like another 20 or 30 seconds. Um I think it's a good time for us to showcase you guys what the uh what the uh sticker shop looks like actually. Let me go ahead and uh um and pull up the

9:33sticker shop right here. Um And while we pull that up, I see some questions in the chat about, you know, the the team functionality and how you share that with other members of your organization and whatnot. Maybe we can address that here in a bit. Um uh just want to acknowledge that I'm seeing those questions come through. Definitely. I'm I'm actually going to post this into uh into the chat right here. Are people able to see that? Are we able to pin

10:01that chat, please? So, people can It's a live website. People can go ahead and play around with it now. Uh I I totally just vibe coded this with Block Bot, full admittedly, and you can play around with it, too. It's like It's like a shopping bag in the middle, right? Uh like this one's kind of

10:20boring. I'm just going to drag this out. Like I don't like it. But then this one right here is looks pretty cute. I'm just going to put it there. And what's important is this is one bot, one person persona, one personality. Each bot has their own personality. You can actually click into it. Let me click into it. And then you can see that this one is called

10:38Thinking. Uh we do a lot of thinking. [laughter] They're still working on the answer. That's pretty funny. And then you can click into the ones that you like to, like this one, nap buddies. Like that's pretty cute. Uh let's see, dizzy. Like this I get dizzy all the time. But play around with it. Let me know what you guys think about it. And I'm going to showcase you guys how exactly how I

11:00built these. Uh let's pause here, right? Let's Let's see if there's any questions or so. Yeah, there's there's some some questions in the chat about sort of the team functionality. And so, um maybe you could talk a little bit around like how you add people to a team as well as, you know, when when I'm asking questions, is that just in my space? Is there a persistent memory with teams? Maybe you can expand a little bit on sort of how that all operates, since I know people are just getting comfortable with this notion of of sort of team bots versus

11:25just personal ones. Definitely. Definitely. So, um it's while we were doing all that, I feel like in AI era, we're just like split-brained all the time. Yes. Like at the same time, Martin actually got back to me, my boy Martin. He said, "Yes and yes." So, he likes the happy blue one, uh and and I think it's good. We'll We'll probably put it on the website in a bit. So, to answer the question, right? Uh how did I add Martin into here? Like let's say today, if I want to add someone else, like uh and let's go ahead and say um I'm going to create uh

12:00create a group chat. And in that group chat, I can add text. I can add Martin, and then I can put Martin in there. And that's how I create a group chat between humans, other humans, as well as other shared bots. Um that's that's one way to address it. Yep. And uh these bots, before they're getting shared, all their memory, all their skills, all their routines, are managed by Martin. And then they're able to be shared to me. So, how Martin configures Dex will be on the fly, like

12:32a push base, that I'll be able to use. Yeah, exactly. And so so you can think of it as, you know, Dex is getting managed by Martin, he's pushing, you know, things like skills or, you know, uh additional context there. And then everyone who has access to that is going to reap the benefits of those as well. Nice. Nice. That's good. Um let's see

12:51here. Let's go back to a slides right. Uh let's go back to slideshow right here. So, you you guys can see that um seems like it's still loading. Uh that's fine. You guys can see that right now we have the designer, which is Dex. That's the one that I just showcased you guys. It's a team bot managed by Martin. Uh and then I use actually Lou, a full stack developer, uh as well as Sam, a UI UX designer. I use these three bots to create the website that you guys just

13:26saw. Go ahead and play around with it. The reason why I'm emphasizing so much on playing around with it is because there's something later in the demo that I'm going to showcase you guys, right? Um and to build this sticker shop, there's actually a fourth one. There's actually a fourth bot that's hidden. If we have time, uh I'll definitely be able to demo this to you guys, but that's exactly how you get your full bot to become extremely autonomous. We'll be able to update the website on its own. This fourth bot is

13:58key. I'll I'll show you guys in a bit. So, let's go ahead and jump into the demo, right? How did I build this website? Let me go back to Grok Bot here. Uh and in this case, I already have my Lou, my Sam, as well as my Dex, the UI designer, the designer, as well as the full stack developer. So, I'm going to do everything from scratch so then you guys can learn along the way. I'm going to create a bot. I'm just going to hit

14:25this plus icon here. Hit create new bot. In this new bot, I can just chat with it and say um you are a UI UX designer. UI UX designer. And I really like and then the website that I really, really like that I got true inspiration from is Duolingo store. Like it looks extremely clean to me. Like you know how I recently there's a lot of sloppy uh sloppy websites. They just have a bunch of subtext, a bunch of text. This one is like the opposite of it. It's extremely clean. I love it. I'm just going to copy this and then say uh I

15:08really like this. Uh and then the goal is to create a um sticker shop that is for Grok bots and here are some assets that you'll be able to use. And then these assets are actually uh assets that um that Martin gave to me in uh Grok bot 101. Uh these six of them. So, I'm just

15:34going to drop them in here. And then what's very interesting here is I'm actually able to connect Grok bot with Cursor. And I'll show you guys right here. I can say something like go ahead and use the cloud agent and launch three models. And then these three models that I want is, of course, Grok 4.7. Uh I'll just use low reasoning for now. And I'm going to use uh let's see here. I think Opus 5.5 is a

16:06fan favorite. I'll put that in there. And let's use Gemini as well. Gemini 3.7. It's It's pretty good as well. I'm going to do low reasoning. Once you're done with everything, go ahead and send me screenshots and screen recording, so I'll be able to see. Go ahead and send that in. So, when we talk to our design design team at SpaceX AI, it's very interesting because this is the new way of coding. Everyone actually just uses Grok Bot today and hook up to

16:37whatever harness that they want to. Today, right now, I'm actually just prompting the agent and say, "Go ahead and launch three different models." Because each model has their own personality, their post-training sessions are different, their their pre-training foundation is different. So, then the with the same exact prompt, you'll get different outcomes. And how do I actually ensure that my UI is extremely polished, extremely beautiful, is to have multiple runs and then pick the runs that I like the most. So, you can actually go ahead and see that right now Grok Bot is went ahead and started studying the assets of the Figma, studying the Duolingo store, and

17:24afterwards, it will go ahead and launch three low-effort cloud agent to be able to build with that. I know there's I I talked about a lot of things. I'm sure I'm Scott is beaming at me. Scott is like, "There's a lot of questions." Yeah, go ahead and shoot. Yeah, yeah, yeah. Maybe maybe first and foremost, too, I think is a little bit of a clarification there. There's some questions coming in the chat around sort of model and where the model's being used. I just want to be clear on the like sort of the prompt there is that we are actually prompting Cursor use cloud agents to go and create these mock-ups

17:57based on each model, and then we're going to make a decision on which one we like best. And so, you know, people are asking about model decisions. And so, just just to be clear, we are actually going and prompting and having them select that within the Cursor ecosystem to deploy those cloud agents. So, Yes. Yes. So, in this case, I'm basically using multiple models to do the same task because I'm I'm optimizing for outcome. I'm optimizing for what my UI will look like and I'm going to

18:24choose the ones I like best. Yeah, exactly. Cool. And then one other question that actually came up quite a few times was you know, I noticed that you're using different bots for, you know, all sorts of different tasks. You have a UX you have a UX bot as well as a designer bot. Why not just have one bot that just controls it all? So maybe you could speak a little bit about specialization and and and why we do

18:47that. Definitely, definitely. And I think that's very interesting because if we think about modeling a startup team, right? If you just have all all the tasks that you want and you you you delegate that to just one person to do it, I don't think that outcome will become great because if that outcome that outcome won't be great because it it if that outcome is always good, then we wouldn't have job titles today. We wouldn't have say UX designer, we wouldn't have front-end developer, back-end developer, and so on. Well, with AI we do see the lines between like some job roles are blurring where back then maybe QA developers, they they only

19:32do QA, but then now we actually see them doing development. Same with front-end engineer to back-end, back-end to front-end. We see that line blurring, but then in my mind, you still will be able to you still want to delegate them separately because they still have their own computer, they have their own context. Context is king in this era,

19:54and they have their own memory as well. Yeah, yeah, definitely. I I mean in my personal experience, it's been this way too. You know, maybe in the future we will just have one bot that, you know, to to do it all, but I think right now being specialized with each sort of task of the process, whether that be software development life cycle or, you know, something that I do in my day-to-day, being more refined with that is actually the best outcome in in my opinion. And and yeah, think of it as, you know, like you would teammates. They're specific to their role and responsibility, and they

20:21become really good at that. So, that's that's been our experience and what we recommend when you're when you're making your bots yourself. I highly agree with that. I would agree with that. So, you guys can see that it actually got back to me. It It went ahead and messaged three cursor agents. It went ahead and messaged the three the three models that I wanted to to ask for. And if I click into these, these actually link me directly into my cursor agents. So, let me go ahead and

20:53and open that up again.

21:03So, you guys are able to see that this is Grok Bot Sticker Shop Grok 4.7 low that is running this. And this is the prompt that my Grok Bot sent to this cloud agent. You can see that there is a brief.md. This is basically advanced prompt engineering. When I think about brief.md, it's similar to plan mode, right? Like in in cursor, those developers that's online

21:30you know exactly what I'm talking about. And it gives assets of the Grok Bot stickers. I only give it a link to Figma. It went ahead and just went to the link, downloaded it, and linked it in the prompt. And it give it like kind of like a goal. And I'm I can assure you that if I open up Opus 5.5, these two probably have the exact same

21:53prompt. Let's see here. Yeah, exact same prompt. But then they're just running three of them in parallel. And you might ask me like, "That's annoying. I need to go on Chrome to look at it." Not really. You can actually see this in Cursor as well. So, this is my Cursor UI. All three of

22:11them are right here. And then you can see all three of the models are just typing away. [laughter] Generating tokens away. They're cooking. They're they're cooking. That's probably the best way to put it. So, while while the the UI UX designer is cooking, let's go ahead and actually build our back end because our back end needs some work. So, let me let me just change this to UI UX so then we don't get confused here. And now let's create another one and call this one

22:44full stack. And in my full stack, I'm going to attempt something that is extremely hard. Bear with me if this goes wrong, but I do this all the time when I'm at home, whenever I'm I'm I'm I'm developing and brainstorming with the model. It's to use the chat functionality with voice chat. Hey full stack, how's it going? Hey, going well. What's on your mind

23:14today? Yeah, so I'm thinking about building a sticker shop e-commerce website. I want something super light. Like I I I I probably don't even need a database. But then I'm I'm deploying it to Vercel, so maybe you can give me some suggestions on what tech stack I use. For that setup, Next.js on the app router with the static export flag is

23:40the move. Vercel treats it as plain static files, so you get zero server cost and instant deploys. For payments, Stripe checkout is the lightest path. Actually, I I I don't think I want Stripe checkout as well. I I I just want like kind of like a mock-up UI, so like Next.js sounds good to me. And then the back end, we can probably just push the the images onto the code base. I don't even want to hook up to

24:07Supabase. Pure front end it is. Stick everything in a public folder or import the images directly. And you can even hardcode your product list as a simple array in a data file. No database, no back end, nothing to maintain. If you want it to feel a little more real without any infra, a JSON file works just as well. Want me to

24:28sketch out the folder structure? Yeah, I I actually think that's pretty good. Do you think you can just kick off a cloud agent and just create a repo and go ahead and and and and and mock up the the UI UX and then go ahead and just send me screenshots and screen recording once

24:48you're done, okay? Spinning up a cloud agent to scaffold the next.js sticker shop repo, mock the UI and UX. And I'll send those over when it's finished. Anything specific you want the design to lean toward? Playful, minimal, bold? I'll I'll tell you later. This is good. Thank you. So that was pretty good. Yeah, that was pretty good. So then we kind of had a 2-minute brainstorming session, you know, like that's this is what I do normally. So my background is I worked at Amazon for about 3 and 1/2 years as a software developer. What I do is before a project I always just call in the PM,

25:26call in the manager and be like, "Hey, let's riff and talk about like how we can do this project." And this is kind of what I'm doing right here with with Grok. So went ahead and talked to full-stack full-stack developer. Full-stack developer said, "Okay, let's go ahead and use this plan." And look at that, already spun up a cloud agent. And if I bet you if I go ahead and switch into my cursor right here, you'll be able to see sticker shop mock-up right here. next.js static is spinning up already. It's spinning up the the scaffolding of what the app would look like and I have the UI UX designer. Oh, look at

26:04that. Someone already got back to me. I'll look at this in a bit, but then we'll have three different choices, right? Three different choices of UI UX and we have the scaffolding of the foundation. We can just put the UI UX onto the scaffolding and then that's pretty much how I built the website. Yeah, it's impressive. I I do also want to say to you there were some questions in the chat, you know, uh you know, we have to go to to cursor to do these for you know, X, Y, and Z. And that that's just particular to this example, right? Of of us going into playing cloud

26:35agents. We also have have tons of different, you know, integration points to all the other systems that you might be working to, as well, too. So, just want to just want to call out that that's particular to this example, but, you know, as as someone like myself, an AI deployment manager, I connect to all the different systems and we'll we'll kick off workflows outside of sort of, you know, the software development life

26:54cycle, as well, too. That's really nice. Yeah. And for those of you that are interested, you can actually go ahead and look at Let me Let me go ahead and go to So, these are actually the ones that I ran specifically for for the website. So, let's let's go ahead and look at Lou, which is the full stack developer. I changed its profile so that there's

27:15like a differentiation. So, I can actually go to its computer, right? And then go go ahead and open this up. So, right now, I'm actually going into the computer of Lou. And this is what Lou's computer look like. And what I want is if I want to go ahead and run terminal commands, I'll be able to do that. You can see that it already ran a bunch of localhost development to make

27:40sure that things are working correctly. I can go ahead and just talk to Lou and say, "Do you mind spinning up the most recent localhost? I just want to see it on your computer, please." And it'll be able to do that. We'll be able to see that it will start controlling its computer. It will start putting commands into its terminal and will spin up localhost and I'll be able to play around with here, or I can just say, "Once you're done, you know, play around with it, record your work, put it in this chat." Multiple different ways that you can play around with. And speaking of earlier, when we used

28:24different harnesses, they'll be able You'll be able to achieve the same thing because the terminal is open right there. Cool. Let's go ahead and jump back to our slides to kind of orient myself where I'm at in the demo. So, the demo looks looks like so and then we were able to go ahead and showcase you guys that we use decks as a designer, Lou as a full stack developer, and Sam as a UI/UX designer to help us achieve this sort of

29:00Grokbot sticker websites. So, again to showcase you guys what it looks like, there's a little bag and you'll be able to drop these these these these little stickers in there. And if you like one, right, you can just click into it and look at what what what it looks like. It's very cute. So, let's see where the progress of the

29:21UI/UX designer is. Oh, yeah, look at that. So, Grok 4.7 actually came first. So, now this is actually the first time every time when there's a new model that drops, I always like to use this concept which is a fan out method where you use the same prompt and then call different models to compare the quality of them. So, looking at Grok 4.7, we can kind of

29:45see like its assets. Looks pretty good. Looks pretty good. And it even gave me a mobile view as well. So, those of you who are joined by Zoom, you can actually use the website on your mobile phone and then it'll be pretty good as well. And it even gave me a screen recording, too.

30:04So, this is something that I didn't do. Keep in mind, I did not record this. This is something that Grokbot did, something that Grok 4.7 did specifically. So, that looks pretty good and let's look at Opus 5.5. That looks pretty good as well. They all look pretty good. So, the reason So, I think right now the outputs I'm getting is actually very

30:32similar to my current shop. Yep. I think the reason why is because I'm actually using the same repo. So, maybe there's some contamination going on. Let me show you guys what uh when I first iterated uh looked like. So, Sam is a UI UX designer that I deployed to make this website, right? Let me scroll all the way up to show you guys what what all of these models look like. So, again, the same thing. I say deploy three cloud

30:58agents. It gave me three of them. And this one here, it's Gemini 3. 3.8 flash work. You can see it's it's it's different, right? It's more more uh vibe coded. [laughter] I had to iterate with it so many times. And then finally, it gave me like a cleaner version, but it doesn't have the shopping bag of dropping it, but it does look like Duolingo a lot. And then uh later on, I said, "Oh, I really like this concept. Let's go ahead and do that instead." And then this is Opus 5.5's

31:31work, too. Like, I really like this hero page. This hero page was very cute. And um it actually gave me the inspiration to So, if I I I believe I believe it's in this Like, look at that. The dragging right there was actually giving giving me that inspo to to to actually make this website drag

31:51and drop. And then it also has this cute little animation that's going on. That's also giving me inspo for that. So, if we go to our current website here and I click on a random one, it has that shimmer. It has that little cute drop. If I like it, I go ahead and do a little heart over here. So, as you can see, if I only use one model with Gemini 3. 7 and Grok, I wasn't able to actually get this effect. It was because I standed out and I was like, "Oh, I like a little bit of this. I like a little bit of that." And

32:27I was able to come up with my uh final product, which is today's sticker shop. Yeah. Yeah. And and and you could even think of a world, too, of like, you know, having a uh a bot yourself even going and reviewing that and giving an opinion on you know design as well too in the future, right?

32:43Yeah, like go and review. Go and review and give me your opinion on it. And so there's there's multiple iterations of this. I also will say that you're getting a lot of love for these stickers. People are asking to order them. And for real. So more to come there I suppose. Yeah, yeah. I think I think if we were to do it in the future it definitely would be like I don't know like 25

33:03cents. Yeah. [laughter] 25 cents per per sticker or free. Well, we'll never we'll see. We'll see. Um so what's interesting is the the the concept I want to anchor on is this AI maturity curve that we talk about a lot. We talked about it in in our 101 session and I'm going to emphasize it again in

33:23our 201 session. In the AI maturity curve we start our journey back in 2020, 2021 when GPT-2, GPT-3 came out. We used the ask. It's like a chatbot, right? Later on it evolved into kind of like a co-work situation. With Claude co-work you'll be able to do certain task. You schedule it and say, "Hey, go do this task for me at this specific time." Now we're coming up to become in the era where we start delegating into our AI bots. This is the maturity curve that we're seeing. Right now we're and and once you have one bot you'll be able to create multiple bots and share it with your team to make it

34:10become a team bot. Right now I would say this website that I created which in total maybe took me two hours in total to make to did the whole mock up, did the whole ordering system. I would say we're probably like right here. Yeah, what would you say? Yeah, definitely. Right here. Yeah. The reason why Oh, yeah, yeah. Go ahead. Oh, I was just going to also say like proactivity has also been coming up as well too. Like I I think that that's a big element of getting up this maturity curve and people probably have seen that in our recent release notes is like, you know, not only just this sort of

34:44maturity of of us us having to use the bots, but also proactivity that we've seen, you know, be a really enhancement to our our our bots ourself. 100%. I I fully agree with that. And I would The reason why I would put myself at 80% 90% at the moment is because we're not at 100% yet. And remember at the first start of the call I said, if we have time, I'll be able to introduce you to one more bot that would bring us up to 100%. And this is the moment. All right, this is the moment to showcase that. That bot that I'm going to create

35:18is called Ed. It's a metrics bot. Uh those of you uh might say, "What would metrics do? Like what what is metrics? Why why would it bring it to to full autonomy, full AI team that's gone?" And here I'm going to explain it. Let's go ahead and jump back into Grok Bot. Ed is right here. It has

35:40everything. Uh and I'm just going to say, "Go ahead and update the popularity canvas for me." So, earlier um I actually in this website, I have tracking in Versel. In Versel, um I grab all the page views. The reason why I keep on encouraging everyone to go on that website and play around with it is because I want to see which sticker is winning. You can see earlier today today at 8:30 a.m., I asked it to do an

36:12update of the popularity right now. At that time, it wasn't public. It was just me. I'm just playing around with it. And I tracked it up this uh this canvas. It's something that's not released externally, so you guys can get a little sneak peek of what it looks like. It's basically a live dashboard that will help you track things very easily. So, when it's just me, there's uh only eight out of 24 stickers that got clicked, right? Number one was party. I really like the party one, you know, it has a little party hat. It has five page views. And then the analytics data was about 2.5 hours. It's hooked up

36:53to Vercel. Just so you guys know. So, let's go ahead and and seems like it's pulling its data right now. Let's go back and look at this. We got we got all the other ones, too. And and at at the same time people can actually place their wager now, like which which sticker would actually win, right? There's like grumpy, dizzy. Wool wedge is pretty funny. I really like that one,

37:18too. Um it seems like the bot is still running. I get it because influx of traffic just came into Vercel. So, I actually did prepare some some a couple of slides to showcase you guys how things work out here. So, how do I actually make my AI maturity curve to 100%, right? Let me showcase you guys

37:40here. Over here, I have an orchestrator bot called Roy. Roy is my primary bot right here. His job is essentially to help me find PMF of these stickers. A lot of times when I when I when I create stickers, I show Martin, and then Martin say, "Oh, it looks good." But only only have two data points. We don't have too

38:03many people. We want to see more, right? So, what Roy can do is Roy go ahead and is is connected to X where I just posted, connected to the website where I can see Vercel, too. And Roy coordinates with multiple different bots. Roy coordinates with Ed, which is metrics that collects data. It's connected to X

38:27as well as the website. And then it's also connected to Dex. Dex, remember earlier? It's a designer for stickers. It automatically links to Figma and is able to create the sad blue the happy blue one. The happy blue one and also alongside with all the other ones that I created in the sticker shop. And it has The reason why it has this very cool animation is because that is a team bot and it's created by our field engineer Martin, extremely talented

38:58individual. And it's connected to Figma, that I'll be able to load in all the assets. And lastly, we also have Lou. Lou just codes. It's a full stack developer. You guys saw the brainstorming session that I had, the 2 minutes of it, talking to it to see which stack that we should use. And it's connected to GitHub so

39:18that it can push code for me. So, why would I want all four of these, right? How do I make them become a fully autonomous AI website? How you can do that is I can go ahead and schedule a routine every 24 hours. And for Roy, which is the orchestrator, I'm just going to go ahead and say, "Hey Roy, go ahead and uh kick off sticker PMF search." What What What that is going to do is it's going to uh talk to Ed, look at all the metrics that's happening. Uh it will have the canvas

39:56that we saw. In that canvas, I'm going to delete the bottom three stickers. Those bottom three stickers gets eliminated. Once they get eliminated, I'm going to ask Dex to go ahead and create three more. And those three will be and will be passed along to Lou. Lou is our software engineer that's going to to update the website and just kill off the last the the three of the worst ones and then put on three new ones. And then we have a reset of 24 hours and we have a fully evolving sticker shop to see what stickers will always forever be be there. And your content will never will

40:38will never go old. Yeah, it's a pretty impressive way to bring like autonomy into something that would be pretty manual work. And you you could also do checkpoints, right, of you know, daily, you know, send me a brief of which stickers are the least performing and and these types of things to keep you in the loop and uh and be aware of sort of what your teammate bots are are doing on on on

40:59their own. 100%. Thank you for bringing that up. Um we've been talking about fully autonomy all the time, but how do we Before we become fully autonomous, we also need human in the loop, too. Um right now I'm painting a very beautiful picture. That's because uh we have expertise. We've done this multiple times and we've done multiple iterations, too. So, within Ed, I actually went through with Ed almost 30 minutes to get down the flow. With Lou, same. Dex, same. And then the full end-to-end loop also took some time to make sure that all the kinks, all the rough edges were were were ironed out, too. So, it's not something that's super

41:41easy that just goes ahead, snap your finger, everything works. Human in the loop is still quite important, but later on you'll be able to take that out more and more. So, let's go ahead and come back here. Uh I think I bought Ed enough time to come back with a canvas. Is everyone ready to see the

41:59Let's see the result. Let's see the results. Let's see. Uh look at that. Piggyback. Piggyback number one, right? And party, yes. Number two. This is yours. Yeah, I know. Cloud nap. Cloud nap is number three. Take third. So, page view and visitors, unique visitors, these are all very nice and these are all pulled from my Vercel. So, what I did was I just have Ed sign in to Vercel on its computer. Uh let me

42:28showcase that to you guys real quick. Come in here, uh open up its computer, come into Chrome, top type vercel.com, and then sign in, and then Ed will be able to see all all all the all the information of the analytics. That's one way to do it. Another way is maybe you have Vercel token, API token that you

42:48want, you can connect with it as well. Um and it's the secrets are not stored in the bot's computer. It's separate so then it's safe. The bots are not able if the bots they they won't be able to leak that information. So, getting back to our full fully autonomy, right? Our fully autonomy is to find PMF. That's most important thing. So, all these stickers, you know, 1 through 21 are all safe. We got a typo over here. There's two 21s. 22, 23, 24, Sprout, Heart Eyes, and Grumpy, unfortunately, we will have to eliminate

43:28them. So, out of three, unfortunately. Very sad. Very sad. So, actually, these three are actually tied, right? Okay. They're both 20s. And I came into So, anyone wants to save them, right? Go ahead and just like spam [laughter] their websites so that they won't get eliminated before I execute it. But [laughter] But these three are are tied. So, then what I told and I ran into this edge case last night actually. So, what I told Roy was if there's a tie case, just randomly select one of them. So, then so then we always only delete three of them. Not not only delete two or delete all five of them. So, let's go ahead and

44:12see this workflow. I come back to Roy and I'm just going to say go ahead and run the PMF skill that I created last night. And please outline and I can send this in and one thing I love about Grok Bot is it's non-blocking so you can keep sending messages and it'll come back to me. Can you actually showcase me what the workflow looks like and maybe draw a

44:40design diagram for it, too? So, you can see that it went ahead and messaged Ed as well as Lou. I I bet the message here to Ed is to is to actually find Okay, PMF round two. It goes ahead and and find a Versel analytics and reply with full list once it's done. So, it's grabbing it right now. So, if you guys want to save any of it,

45:04this is this your only chance. I went ahead and messaged Lou as well. So, the message to Lou is over here. So, you guys can see that this is the the workflow that I had. This one here is actually not not as good as I like it to be. Earlier when I tested it, it looks a lot better and I can kind of show you guys what it looks like. Basically, I'm right here. I go ahead and talk to Roy, which is my primary bot. I said go ahead and find PMF. It's going to talk to Dex, which is designs new stickers, talk to Lou, that swaps out the stickers in the

45:39code base, as well as Sam, which goes ahead actually doesn't have any part in the PMF loot, and then it talks to Ed to read the Versel analytics per page. Once it gets that, goes ahead and talk to Lou talk to Dex, grab the new designs, talk to Lou, deploy to Versel, as well as you use Grok 4.7 low to go ahead and do that. It's fully autonomous, runs it 24 hours

46:06once every 24 hours. Yeah. So, so you can see that right now it's it's just going to run. I don't have to do anything anymore. And then our website will get updated and then the the bottom three, unfortunately, will get eliminated, but we'll get new ones and that's how you get a self-evolving website, right? So, then

46:27your content will never go stale. That's awesome. It's an incredible example of really bringing that full loop and and and allowing your bots to work for you and and and freeing up your time to be more like strategic with your work. I just want to give an update here. Some folks in the chat are reporting that Joseph has increased his follower count on X by 15. So, I just want to just want to call that

46:49out and say congratulations to Joseph. Thank you all for 300%. [laughter] Also, to really tactically like some folks were asking about the the the ability to chat with your bots, which is an awesome feature. I recommend you all start to do it. It's kind of change the way that I've worked with my bots transparently and and the way that you do that is just on that bottom

47:08right-hand corner of the chat interface. What Joseph was doing was just dictating and and speaking to the bots to be able to actually prompt as well, too. So there's a little bit of difference there of dictation and then, you know, being able to communicate live with with your with your bots. Definitely. Definitely. That's a that's a good way. I think I think mostly we covered that in the 101 session, but I

47:29would love to walk through that, right? You'll be able to type and just say, "Hey, I what's up?" You'll be able to use this dictation function and you can click on it and say, "Hi, what's up?" And then this underneath the hood we use Grok API for voice and you can see that it works very well. And earlier in the demo I can talk to Grok bot, which right now I'll I'll I'll remain silent for it, but you can talk to it back and forth with very little latency. I really like

47:55that. Yeah, it's really really impressive. I know that there was comments about it not sounding like a robot, so that's impressive. So shout out our our our Grok voice team. [laughter] Yeah, they're they're doing an awesome job. It's very cool. So, lastly I just want to go through some some recaps of what we learned today before we jump lastly into Q&A. So, what did we learn in 201, right? First thing, most important, every bot has their own persona, they have their own memory, they have their own skill, they have their own routines. You can set up

48:29routines just like what I did with Roy. Every 24 hours it purges three of the bots, replace it with new ones, so then we'll be able to complete keep on iterating and self-evolving loop to find PMF. And then we use cloud agents. All these code I did with designing the UI/UX, scaffolding the foundation of the website, all of that is using cursor

48:54cloud agents. I use a skill, that workflow that you guys saw was created with Roy where it coordinates with multiple bots. And lastly, team bots. This is extremely powerful. We'll be able to see that more and more in the future. Imagine if you put all the expertise of creating a sticker into decks, right? Like what Martin did. Um it's extremely good at creating uh designs for Grokbot logos, cute designs that we saw all 24 of them, and now it's going to be 27 of them. Um you can share that with your team. We have about 100 people, 150 people in our field engineering team. We'll be able to

49:38share that with all 150 of them. They can create the stickers whenever they want, and we can also use uh Martin's uh sourcing uh Penny uh to help us uh create the stickers for us, too, if we publish the team bot. So, think of some ways that you can benefit your team uh in the future. Uh we see that kind of like You know how like 6 months ago

50:02everyone's writing skills? Yeah. Uh 6 to 8 months ago people are writing skills and publishing it and showing it to Markdowns were everywhere. Yeah, markdowns are everywhere. Now we're seeing bots are getting shared everywhere, and only the best ones survive. And that's what we see within Cursor internally, too. Engineers are creating their own custom bots, sharing across all different teams, finding PMF within their own company, so then their bots can become better and better and become almost like a teammate within

50:31you, as well. Yeah, really really impressive stuff. Um I I did just want to call out, too, um yeah, sorry I'm going to interrupt. I I have some There was some chatter uh in the chat about sort of uh the emails functionality that I know that uh just got released while we were while we were live, I think, or maybe a couple minutes before. Uh but there's some really cool new functionality that you can actually have your Grokbot go and claim its own email address so that it can have its own sign up for different services. It can actually go and, you know, join uh meetings on your behalf or book meetings

51:03on your behalf. So, it's really cool functionality. I definitely recommend people go and just prompt your Grok Bot to go and claim an email itself so it can go and do that. It's really really really as simple as that. Just go to your Grok Bot and ask ask it to claim an email for you. That's actually super helpful because let's say like tonight if I want to go on if I want to go to Foreign Cinema for dinner. It's like a pretty popular place in in SF and then if I want to go ahead and do that it needs resi and it needs a email. So, I can just go ahead and say,

51:34"Hey Grok Bot, do some reservation here. Use your own email." It'll do it. Yeah. Yeah, exactly. That's so cool. I love that. I love that. I'm definitely going to use that then. Um It's also early days. So, go and claim a nice URL I don't know if Scott has it taken. Yeah, I'm going to [laughter]

51:54go and grab it. And yeah, that's it. So, thank you guys so much. Thank you guys so much and back to questions but I'm I'm sure the questions are trickling along. So, I'm going to stop sharing my screen here and answer a couple of ending notes. Yeah. Yeah, we have a couple questions while we have a couple minutes. So, so people were asking about the fan out method and and sort of your perspective on that, you know, weighing the balance of token consumption versus you know, doing a bunch of fan outs and I don't know if you have a perspective yourself

52:22on someone who does this more regularly. Definitely. Definitely. I I do have a method for that is when I'm doing fan out, I don't immediately do fan out right away. I actually do one particular model that I'm familiar with. For me, my case, I'm very familiar with Grok so I know how to prompt it, how to interact it. I kind of know what what it response it'll get out from me and I build a workflow of what that prompt will look like what I'm looking for. And once I have that, then I go ahead and fan out because if you just mindlessly prompt and just do say say say things that are not

53:07specific to the model. It could be interpreted in different ways. So then it will cause you token inefficiency. Yeah. I think that's important strategy. I think I would say for for anyone on on thinking about the projects that you're going into, you know, if you're most likely everyone here is trying to be token like conscious and so think about the planning stage of it all before you go all in of of things. And so I think that that's really important advice

53:32across the board. Yeah, and that also reminds me of earlier remember I was talking to full stack guys. When I was talking to full stack Lou with the voice back and forth, it actually thought about token token consumption for me. It said something like uh we should probably use Vercel next js so then we actually stay on free plan. So lots of times when you keep talking to the model, the model understands you and know what your preferences are. I

54:06definitely wanted the free plan. Yeah, definitely. It then knows my preferences and will be able to keep getting better towards whatever you want. If you're more performance driven, I'm sure the bot will start understanding that as well. Yeah, yeah, there's sort of an element and this actually goes back to a question around, you know, things like the primary bot or a chief of staff bot that we probably went over in a in the 101, but if you missed it if like, you know, what are your preferences? Of course the bots are going to learn yourself, but there's always a great opportunity to be very explicit with

54:38sort of how you want to operate and and how you want to go about your daily tasks. I definitely recommend taking that time at the beginning as you're getting started to to really think about that and instruct your bots accordingly. That's very true. Yeah. Um Cool. I know we are just about at the end so going to address the biggest question in the room which is will this be shared out? And I do want to

54:59emphasize yes, it will be. Give us a couple of days and it will be out on our workshops, but Joseph did a lot of the heavy lifting today. so thank you so much Joe for his also impressive sticker shop and yeah, thank you all for joining. We had a we had a huge crowd today. So really appreciate your time. Sorry if I didn't get all to your questions. There was a lot of them, but have a have a great rest of your week and weekend everyone. Thank you

55:21thank you for joining today. See you guys. Bye.