← AI guides
Grok Bot 101: how to build a team of AI bots
Grok Bot 101 Play video Playing loads the video from YouTube (youtube-nocookie.com).
Subtitles: on
BG EN
Turn on JavaScript to play the video with synced subtitles, or watch it on YouTube. Watch on YouTube
Video: “Grok Bot 101” by Grok, from the official @Grok YouTube channel. All rights belong to its owner. Watch on YouTube ↗ · @Grok
Summary
In this official workshop, published on the @Grok YouTube channel, Cliff from the AI deployment team at SpaceXAI and the engineer Martin present Grok Bot – a tool that, in their words, is the first true AI colleague you can delegate work to.
Martin starts with a short history: chatbots around 2023, then connectors and MCP, and now bots that work asynchronously away from your laptop. The two core ideas of Grok Bot are persistent memory – you can talk to the same bot for months – and each bot's own computer, which it uses to work with the internet and apps almost like a person. As an example, Martin shows his personal bot Bobo, which keeps track of new clothing drops, fantasy football and tennis tickets, and books trips.
In the demo he builds a team of three bots. Dex designs Grok Bot stickers following guidelines in Figma. Maya creates a Google voting form with a QR code after Martin shows her how with the "Teach a Task" feature, and she saves it as a skill. Penny looks for print shops in New York after Martin gives her the task by voice chat. The bots message each other, the audience votes live, and finally Bobo is promoted to chief of staff of the team in a group chat.
Martin sums up: bots can access tools even without MCP, they remember how you work, and in the settings you can set rules for when they must ask for approval first.
Full transcript
Click any line to jump to that moment in the video.
0:01 Hello, my name is Cliff.
0:03 I'm joined here with Martin. We are going to be, well,
0:06 Martin is going to be running the 101 today.
0:09 I will be moderating the chat, so please post your questions,
0:13 and I'll try to respond to them.
0:15 As many as I can, I probably won't be able to get to all of them,
0:18 and I will surface what is most important to Martin so we can cover them
0:22 live for everyone in the audience where relevant.
0:25 I work on the AI deployment team here at SpaceXAI,
0:28 so I work with tons of customers,
0:31 to help them get the most value out of the platform,
0:33 everything from workshops to workflows, things just like this.
0:36 And I'm joined by Martin, who I will pass it off to,
0:39 who is a field engineer here at SpaceXAI. Thanks for coming, everyone.
0:44 Hi, nice to meet you all.
0:46 Thank you, Cliff for the introduction.
0:48 By the way, Cliff is awesome. So be kind to him in the Q&A over there.
0:51 I know he'll be locked in and slammed the questions.
0:54 But nice to meet you all. I'm a field engineer here on the SpaceXAI side.
0:59 I originally came from Cursor.
1:00 And so now as we expanded the actual ecosystem of products
1:04 that we help educate and give the best tips and tricks
1:07 that now includes the Grok main, Grok APIs, and now also Grok Bot.
1:12 So everything Grok related, all fun there.
1:16 But more so extremely excited to hop into a 101 session on our latest
1:21 product, which is now in the general knowledge work.
1:24 And so before really kicking off here, I would love to know if you could put a.
1:30 rating in the chat on how well you know Grok Bot where zero is
1:36 this is like your first time actually seeing or hearing what it's going to do
1:40 to five you already have an AI workforce that is already doing some crazy
1:46 things behind the scenes with Grok Bot.
1:49 So if you could put a zero to five rating in the chat,
1:51 that would be awesome to see there just to get a gauge of the audience we have.
1:56 Also, Cliff, I am not able to see the Zoom chat,
2:00 so feel free to let me know what numbers we got here.
2:03 I'm seeing the full spectrum of ratings.
2:06 A lot of zeros, a lot of folks above five,
2:11 really every number.
2:14 Any 1,000 in there?
2:17 Nobody's put 1,000 yet. Okay.
2:20 If you are at 1,000, hopefully there is something you can take away in this 101.
2:25 And as well, we have a 201 coming up with Joseph, which amazing flowed engineer.
2:31 Then you can learn a lot more stuff there.
2:33 But this would be more or less the basics of how to use Grok Bot.
2:36 Some context on me.
2:37 So it was a previous software engineer used Cursor from 2023,
2:41 that IDE agents window to like, you know,
2:45 2025 of launching cloud agents during team meetings.
2:49 I hope my old manager isn't in this workshop hearing that.
2:51 But now more so, really excited to showcase this because it is our.
2:55 first tool that is for everyone across the board.
2:59 So even all the segments, I'm sure you saw the Grok Bot Galaxy,
3:03 if you're going to go to market, if you are in product,
3:06 if you're in engineering,
3:07 or even just a hobbyist and you want to be able to have a simple interface,
3:11 but really leverage power of AI to do real work.
3:15 This is the workshop for you. So I'm glad that you joined us today.
3:21 So just for some housekeeping, this is being recorded.
3:25 And if you have any questions,
3:27 I would recommend you don't put them in the chat,
3:30 but you put them in the Q&A section,
3:32 just so Cliff will be able to actually manage and answer whatever questions
3:36 may come up as well too.
3:38 And we can showcase in the agenda.
3:41 We can have like sections of if questions arise with common themes and ideas.
3:46 Cliff can go off mute and then let me know. We can handle that.
3:49 And so it could be beneficial for everyone to hear that across the board.
3:53 Awesome.
3:55 Just a level set on the agenda,
3:58 really want to showcase the history in the last couple of years of how AI
4:02 has transcended, how we use it today,
4:05 to why we even brought Grok Bot into market.
4:07 And I think there's a really fascinating story there.
4:10 And then we'll showcase Grok Bot inaction, so showcasing the platform.
4:15 And then we'll sprinkle every bits and pieces of potential questions
4:19 if there is room.
4:20 But more or less, we'll have 10 minutes at the end just for Q&A.
4:25 if there's anything that arises throughout the entire workshop
4:28 that you have a question on.
4:32 Great.
4:33 And so before now going into Grok Bot,
4:35 I think talking about the history and the way that we've actually brought
4:39 this product into the market is important.
4:42 And so let's start off with where we had AI a couple of years ago.
4:46 So around the 2023 time, this is where the concept of chatbots came into play.
4:52 And I remember distinctly, I was in my living room, I was on the couch,
4:55 asking Chad GPT.
4:57 you know, like what is the meaning of life?
4:59 What is like the weather tomorrow?
5:00 I'm just asking crazy questions to see what exactly can come back to me with
5:04 and extremely surprised at how conversational it actually was.
5:08 And moving more into a year where connectors and MCPs came into play
5:12 where we're able now to give AI a specific, you know,
5:18 action or task.
5:19 And we can connect a Figma MCP to now work in a turn-based style to complete
5:25 a slide deck, whether that is in a couple of hours and actually go back
5:30 and forth with it to do that. Now,
5:33 where we've seen Grok Bot come into play is being able to actually take
5:38 that execution off of your laptop,
5:40 move a bit away from that turn-based language of talking with your AI,
5:44 and now actually being able to have a colleague, have a teammate,
5:48 that you're able to trust a job and have it actually delegate this outcome
5:52 to it.
5:53 So Grok Bot is something that before we even launch it into production,
5:57 I believe last month, Cliff let me know if I'm off base there,
6:01 but we were using Grok Bot internally across our entire work from engineering
6:06 to go to market.
6:07 And so like we're talking to specific Grok Bots like one conversation for
6:12 months on end and is able to actually learn.
6:14 So when we delegate an outcome,
6:16 because we essentially onboarded it as if it was a new hire,
6:19 it was as easy as being able to now just pass future work.
6:24 And so moving forward up as maturity curve now is where we see at SpaceXAI,
6:29 the evolution of how AI works,
6:31 where no longer we're working with one teammate to delegate work to.
6:36 But now we have a company or cohort of teammates.
6:39 So I view it less as agents warms.
6:41 I view it more as like an entire AI workforce that is able to act on
6:45 your behalf so that ideas that you might not have thought were possible
6:49 before are now meant possible because Grok Bot exists.
6:53 and because we actually have teammates that are able to talk to each other.
6:58 And so this is now moving into, right,
7:01 like Grok Bot and where we see in today's market,
7:04 this is a first ever true AI teammate and colleague that you were able
7:08 to delegate work to.
7:10 And so if you notice on the screenshot here,
7:14 it should feel familiar and simple as if it was iMessage.
7:18 The main premise was that we wanted this tool to be extremely easy,
7:22 to be onboarded onto regardless of if you are in a profession,
7:26 if you're a hobbyist,
7:27 if you know, you're even a parent and you want to see cooking techniques
7:31 and recipes to use for tonight's meal.
7:33 This should be as simple as texting a colleague,
7:36 right, where you have all your bots on the left-hand side.
7:39 We have the sales outbound.
7:41 We have the inbox manager.
7:43 And this is where we'll briefly touch on the 101,
7:46 but even have a chief of staff that is able to talk to these different bots
7:51 on our behalf.
7:53 In the middle section, going from left to right here,
7:55 is just the single point of conversation where we are able to talk to
7:59 our teammate. So I know in today's tools,
8:02 you have to create a new conversation, hit that plus button every single time.
8:06 You have a new task.
8:07 That experience feels a little bit less cohesive when we think about
8:10 actually how we talk to teammates and colleagues and how we distribute work
8:14 to others.
8:15 This is now able to have that memory as we talk with it over time
8:20 and is able to already now do work and even without prompting it.
8:25 So if you realize there is a little snippet here is called created the
8:29 sales outbound routine.
8:31 This is where on the right hand side.
8:33 We can see how our bot is actually operating asynchronously even without
8:36 us prompting it.
8:37 So name, description, but also the actual routines of every morning at 8 a.m.
8:43 If this sales outbound bot is doing PG,
8:47 automatically send to everyone in our queue those drafted emails of how
8:51 I actually taught you to do that.
8:54 And so in short here, right, Grok Bot is that true first teammate.
8:58 And I think the two product decisions that we create with the Grok Bot,
9:04 first is being able to have persistent memory.
9:06 So actually being able to talk with this bot until the end of time
9:10 and it knows exactly how it operates.
9:13 It knows the type of work that we do.
9:14 It's handling a very specific role for that job.
9:17 you are typically used to doing,
9:20 whether it is a rote or mundane work or even meaningful work
9:23 that you want to pass it off to.
9:24 But secondly now, the fact that we actually give each Grok Bot a computer.
9:31 So if you think about the current tools today,
9:34 when it does, the execution does it on your laptop.
9:36 So if I'm creating a slide deck,
9:38 the moment I actually want to go out to lunch, close my laptop,
9:41 all of that just gets absolutely paused, when in reality,
9:45 if we actually have a real 24-7, ASync agent or ASync teammate,
9:50 it has its own computer that is able to actually do work off of our laptops,
9:56 or sorry, out of our laptops, and is now able to do long-running work there.
10:00 And even more important, because it has a own computer,
10:02 it's able to actually use the internet very similar to how we use it.
10:06 So you'll realize that there are a lot of MCPs and connections
10:08 that we have out of the box, which will showcase in the marketplace,
10:12 whether that is pulling at Lassian tickets or documents
10:16 or whether it's pulling conversations from Slack.
10:20 We can also just have our bot, go to United Airlines.com,
10:24 and actually just book a flight travel for us because why would United have
10:28 an MCP or an API for that matter?
10:30 If it does, I would be extremely interested.
10:33 But for everything else, even internal dashboards,
10:36 you're able to have RockBot just point at it and do the work specifically
10:39 how you already would.
10:41 And as a little teaser,
10:42 we can even teach the bot how to use a computer with a teach a task.
10:46 feature but that would be more later coming up take a little pause here i
10:53 am curious cliff i know you're locked in with messaging there the q-nas is
10:57 there a bot in mind that you've used in the past
11:02 that you've wouldn't say aha moment but like you is extremely useful in your
11:06 day to day
11:07 yeah i would say um so i work like i mentioned i work with a lot of
11:12 customers on helping to like deploy SpaceXAI products into their company
11:17 to help their employees work.
11:21 And so what's hard is finding,
11:24 you know, opportunities and risks and things proactively.
11:29 And I have this bot that I named driver that's like my customer driver,
11:33 if that makes sense, like the driver of opportunities and getting things done.
11:37 And it helps me to remember things that I've dropped on the floor
11:40 or surface opportunities for like where a customer is having a problem
11:44 and we have something or a tool or a feature whatever
11:48 that could help them things like that the type of work
11:51 that you know you could spend all day Googling and looking on you know what
11:54 are they posting on LinkedIn or their blogs or whatever and you could never
11:57 find everything and this bot is that proactive like opportunity searching
12:03 bot to help me do that with my customers so it finds finds things on my behalf
12:08 I think that's that's my favorite type of use case for Grok Bot is
12:11 like proactive monitoring
12:14 not just doing work you already do,
12:16 doing things you couldn't have done before without a helper.
12:19 That's awesome. Thanks for sharing that. Yeah, we even have,
12:22 I've seen, I'm not sure if it's rolled out to everyone just yet,
12:24 but we have a primary Grok Bot where if in this interface you have the
12:29 sales outbound, chief of staff, inbox manager, we now, like,
12:33 from a Grok Bot experience,
12:34 we make one their primary and it's able to actually keep tabs
12:39 and everything across the board of what we're doing. And to your point cliff,
12:42 I remember rescheduling a calendar invite with one of my teammates,
12:46 but didn't actually reschedule it.
12:48 So I woke up to Grok Bot already scheduling that for me
12:51 because it knew that I was going to forget.
12:53 And it was in fact correct.
12:55 I did forget. But I love the name driver.
12:58 I wish I actually named my Grok Bots in a way that makes me remember what
13:02 they actually do. I name it very different names and even cute name.
13:06 So what I want to showcase with the two principles of the persistent memory
13:11 and computer use, I want to showcase with a story with a bot called Bobo.
13:17 So Stilmore to Cliff, I also work and go-to-market.
13:20 So I have all of these workbots that actually help me prep for all of my calls.
13:26 It gives me a rundowns with all the accounts, even account usages,
13:29 all things like that.
13:30 But Bobo is actually far and away from all of my workbots.
13:34 It's actually my number one go-to personal bot,
13:38 and it covers everything outside of my work life.
13:41 And for those that know me in this call,
13:43 I think there's several people that have told me, hey,
13:46 I'm going to be joining no pressure.
13:47 They know that I am a bit of a workaholic in the sense that I even like to
13:52 work into the weekends,
13:53 and I have the systems and routines of being able to wake up early,
13:57 buy my overpriced coffee in New York City and then feel productive for
14:01 that day as a system.
14:03 But with that being said,
14:04 there are things that are meaningful to me that I want to be able to make
14:07 sure that I keep tabs on and then put effort into,
14:10 whether that is my two lovely parents, my two older brothers,
14:13 or maybe even booking adventures so that I'm able to see the world
14:17 and force that free time on my calendar to actually enjoy so
14:21 that hopefully I can have wonderful,
14:23 innocent stories for my grandchildren to tell one day.
14:26 but this is where Bobo comes into play.
14:29 So what I'm showcasing is my mobile app.
14:33 We have the desktop version, which is what I primarily use to work with.
14:37 But if I am on the go, on the way to work,
14:40 or even just walking throughout the office,
14:42 I don't want to handhold my laptop open while it works.
14:45 But I use a mobile app.
14:47 It's very similar to iMessage.
14:48 It feels familiar to me, yet really powerful as a tool.
14:53 But so in terms of persistent memory,
14:56 Bobo is able to actually remember and keep tabs on everything important in
14:59 my life outside of work.
15:01 And so very simple use case,
15:03 you can tell already that there was a routine at 907 a.m.
15:08 Bobo told me that Kith, which is a street apparel brand,
15:12 told me that there was a drop yesterday that my brother might like
15:16 and even recommended me three different apparel items with their prices.
15:20 So as you can see, the Georgia Romani is a $3,000 jacket.
15:27 I don't think, I think Grok Bot thinks that money grows on trees,
15:30 even though I love my brothers. So I'll say, I love my brother,
15:34 but not $3,000 so that it actually knows moving forward,
15:38 that maybe we can have actually a price range that makes sense for me.
15:43 So it looks like the disconnector went off there,
15:45 but you can showcase Grok Bot or Bobo with my desktop.
15:50 to where if we notice it has a couple of routines that it launches whether it
15:54 is looking for that kith drop whether it is all my fancy football leagues
15:59 um seeing of injuries across the board of what i have to do um
16:03 or even like my parents love this tennis player yala uh who crushed in the u
16:08 s open and they want to make sure that they actually see her
16:11 when she comes in new york city so i want to be able to surprise them
16:13 with tickets the moment that that does happen um hope my parents also
16:17 aren't watching this or else that surprise be ruined there but This is more
16:21 or less where Bobo has a memory to know over time specifically how I
16:25 actually operate.
16:27 And even more on the travel side, it already knows allegiances, my hotels,
16:30 my airlines, and it's able to actually book entire itineraries for me.
16:34 So even legs where I have multiple different cities,
16:38 a different destination location, it's able to do that because Bobo,
16:44 for better or for worse, I've given it its own computer.
16:47 so we can see in the right-hand side here that if you open up the computer
16:51 that again is not on mine but Bobo's virtual computer we can see it's
16:56 already looking up hotels to book for me Atlanta is a sample location I'm
17:02 not going there but this is where it's looking at the different flights
17:06 and hotel accommodations able to actually book an itinerary for me
17:10 and it's as simple is having a one-step button of saying confirm booked
17:15 looks good so everything outside of work that might feel mundane
17:21 or tedious to do i have bobo handle everything behind the scenes
17:24 and it's made my life extremely easier on that front there
17:32 awesome and so promise you not this is literally the last slide i don't want
17:36 to bore you all too much with just content before actually moving in to the
17:40 demo itself but just to recap why Grok Bot it's as simple as messaging
17:45 a colleague with iMessage.
17:48 And the best part about these colleagues is that you're able to give them
17:51 a hundred different line items and they actually won't be frustrated with
17:55 you for doing so.
17:56 And secondly, they're always on colleagues.
17:59 So the moment that I actually log off for work,
18:02 it's able to still do these long running tasks behind the scenes
18:05 and is able to give me a summary of what it did the next day once I actually
18:08 do log back in. And so it has all the tools that we do.
18:12 This differs on the actual segment. that you're in, but as well,
18:17 you're able to actually finish the work because we were able to onboard it as
18:20 if it was a hire.
18:22 And what I actually love specifically as well,
18:25 which is more of a recent feature,
18:27 is the fact that we can share these templates of bots to others as well
18:32 to find value immediately.
18:34 So if I am able to actually onboard Bobo and knows exactly how an
18:39 assistant should operate,
18:40 someone else is able to leverage that power almost immediately by getting
18:43 this imported bot template here.
18:47 And lastly for resources,
18:50 what we can showcase is Grok Bot Galaxy was a three-day session
18:55 that was ran back in September.
18:57 This is if you are in a specific segment for engineering, for PMs,
19:03 for go-to-market, even for founders on this call.
19:06 There's 35-minute videos of very bright and exceptional people talking about
19:10 how they actually use Grok Bot.
19:12 And the cool part is we even have Grok Bot guides here,
19:17 where we're able to get a step-by-step play-through of why they create
19:21 a specific bot and being able to actually import a bot.
19:25 So I know here that for the Grok Bot for SDRs,
19:29 there's actually a chief of staff bot called Josh Allen,
19:34 where you're able to actually import this from the marketplace
19:36 and already use in your team if, you know, PG is something of used to you.
19:42 And so maybe if we have enough time,
19:44 we will get to the chief of staff portion of actually how to manage team of
19:47 bots and what that even means, where I can feel a little bit abstract.
19:51 But we have a marketplace we're able to actually get these templates out of
19:54 the box and something that the 201 will cover more so.
20:00 Lovely.
20:01 So let's move into now meeting the team what we'll actually work on today.
20:05 So the main objective is I want to give a workflow where we're using a
20:11 similar tool, whether it's Figma,
20:13 to actually create a Grok Bot designs so that Maya as a community bot is able
20:19 to create Google Form, create a QR code, able to do that using the computer use.
20:25 And then everyone in this call is able to participate by actually scanning
20:28 the QR code and voting on the best specific Grok Bot piece
20:32 that we do make, in this case, it's stickers.
20:34 And then Penny will actually find vendors that's able to print these
20:38 stickers out and distribute them to everyone across the board.
20:42 So this is the workflow of being able to actually have a team delegated to
20:46 their own specific roles where we have decks as a designer,
20:50 Maya focusing on the community aspect,
20:51 and then penny for procurement and sourcing.
20:55 And again, if we have time, I put a question mark here.
20:58 There's a fourth mystery bot that would be able to potentially work in
21:02 this workflow and make this process simpler for us as we actually teach
21:06 these bots to do their different jobs here.
21:11 Awesome. So let's move now into the demo portion of this session.
21:20 I'm going to load back up our interface. Again,
21:24 what I want to highlight is that
21:26 all these different colleagues or bots on our left-hand side should feel
21:30 very familiar to iMessage.
21:32 We're able to click into one and just text in natural language.
21:36 We even have Grok voice,
21:38 which is a very simpler way to actually talk with their bots and
21:41 I know that, like, we even speak with founders who have Teslas
21:45 or even in their cars on the way to work,
21:47 have Grok voice just turn through all their emails and even route to the
21:51 correct person on Slack if that email needs to be escalated. Like,
21:55 extremely cool use cases of just using voice chat to handle all of your
21:58 work across the board with Grok Bot.
22:00 But really getting started here,
22:02 if this is your first time using this interface,
22:04 it's as simple as going to the plus on the top left hand side here.
22:09 This is where we can do create new bot.
22:13 And what we'll notice is that it will say new bot is working.
22:17 It would give us a simple questionnaire of exactly what it will be designed
22:22 to do. Excuse me.
22:25 And so in this case, our team is decks, mine, penny.
22:30 We can just start off with creating decks and what that looks like is we
22:33 can click the new bot at the top here.
22:35 It will pull up our right side panel where I can click on new bot
22:39 and actually name this teammate to be. Dex.
22:44 And then for the add a label,
22:46 this does not actually change the outcome exactly of how your bot works,
22:51 but is more so if, you know, Cliff has driver,
22:54 which actually makes sense with what the bot does.
22:57 I have Bobo. I might forget what it is.
22:59 So I have it as my life stabilizer.
23:01 In this case, Dex is my designer.
23:04 So I can have this label here.
23:06 And so now it's as simple as just prompting it in plain text of exactly
23:10 what the role should be.
23:11 And so on the left-hand side,
23:12 there's different sections we have unassigned bots this is ways of
23:16 just grouping bots of if we are eventually going to move towards a team of
23:20 bots being able to section them and know what they're all working on and
23:25 so i've actually already before just for the sake of time created all the
23:29 three bots i would be working with today so you have maya penny and decks
23:34 and what i'm going to do is highlight them double click
23:37 and do move three bots into a new section Notice and moved into the top here.
23:45 And I will just say,
23:47 Grok Bot sticker team,
23:52 just to get started here.
23:55 So let's start off with specifically with a community.
24:00 So we already know that we want to create a Google form
24:02 and we want to be able to have a QR code so that everyone on this call is
24:07 able to actually vote on the number one rock pot sticker,
24:09 which hasn't been created yet, but we can just focus here specifically.
24:13 So it's as simple again as, as,
24:15 talking to Maya in plain text of what exactly the role will be
24:18 and what we should work on. So Maya said, hey, Maran, good to meet you.
24:24 Nice to meet you too.
24:27 Your role is to create Google Forms so
24:31 that the audience that is watching us right now
24:37 is able to vote on a specific Grok Bot
24:42 sticker that we,
24:47 should end up following through with printing out.
24:54 Your sole job here is just to make Google Forms,
24:59 and I will showcase how to make this for you.
25:07 So now this is where, like, a lot of people that we talk to with Grok Bot
25:12 ask, you know, how do I actually get a bot up and running? What if I'm confused?
25:16 If you ever are, it's very similar to,
25:17 Cursor if you're an engineer of just being able to ask in natural language
25:21 what exactly that you hope to accomplish it can even lead you to the
25:25 correct path of of that angle there so got it i'm in google forms bot for
25:30 the sticker vote pulling up how i build ready when you are
25:34 so what i'll first say is can you open up google forms in
25:40 your computer so i can take control
25:48 and so i'll click on maya and we the details which is the routine
25:53 specifically that Maya will do so again routines is instead of me
25:57 actually prompting Maya to do a specific task we're able to set those
26:03 routines whether weekly monthly or even daily or even hourly a minute of
26:08 when it fires off and does a specific role so all that was showcase here we
26:13 have a library so once we actually work with Maya whether it's photos
26:16 or videos it will showcase here and then computer is where, again,
26:22 we're able to see Maya's screen and actually what Maya is working on.
26:27 So I can click on the screen here.
26:30 Looks like Forms is loading on the computer.
26:33 I'll hand it off to you soon as this up.
26:35 So we can see that Maya went directly to Google Forms in the computer.
26:40 And so what I'll say now is I will take control and actually already knew
26:46 that I wanted to do it.
26:48 So I can say take over now.
26:50 notice how it prompted me.
26:52 This is sort of the going back and forth with the Grok Bot here.
26:56 And so I'm going to say,
27:01 skip this step.
27:04 Let me
27:07 show you create
27:13 a form, BRB. So I'll go back in the computer.
27:17 This is where we have a feature called Teach a Task that I teased so
27:21 that Maya is able to actually see how I do a specific flow.
27:25 In this case, it could be a bit trivial that's a Google form
27:28 but this is across everything whether it's an internal dashboard a platform
27:33 that you want your Grok Bot to point to and actually do work for you
27:37 whether it's many different inputs and even typing in different things
27:41 from different sources able to do all this for you can teach it directly how
27:45 so i'm making a new form in its computer
27:48 notice that it's currently maya's watching and learning let's help up the
27:52 hand side This is where I will say,
27:58 practice,
28:00 Grok Bot sticker votes.
28:04 Form description, allow the audience to vote on
28:10 a sticker.
28:13 And then this first one will just be what is your favorite color?
28:18 So this form actually won't be sent out in a QR code.
28:21 This is more just showcasing to Maya.
28:25 how to create these specific options here i'll say blue i'll say red make
28:32 this required and now the second point to this piece is then creating
28:36 another option where we will use an image so here i have google image
28:42 attached to my account or google drive for example for this matter
28:47 and i'm just going to input this here
28:52 and then type what thicker is your favorite
29:01 And so this is now where we'll make sure that my nose,
29:05 this will be a multiple choice.
29:08 You can say option one,
29:11 option two.
29:12 And so now I'll make this required and now go into publish.
29:18 Make sure that anyone on this call is able to also access this.
29:22 So anyone with a link, perfect.
29:24 And then we will say publish.
29:27 And then we will say
29:29 pause the video there.
29:32 So if we notice, what I will say to as well is, by the way,
29:38 once we eventually get to the sticker portion, which is later,
29:45 you will create a QR code.
29:49 So the audience can use the Google form.
29:55 Number one thing to note is that once I actually finished that recording,
29:59 it was able to tell me, what skill would actually use to teach our bot.
30:04 So for context, I could have told Maya to just create this Google form.
30:10 It would have went through it,
30:11 looked at the HTML divs to see what potential options is what Martin's
30:15 looking for.
30:17 When we think about tokenomics and being able to actually drive our bot in
30:20 a behavior that makes sense to us, this is where skills comes back into play.
30:25 So for us to be able to actually showcase,
30:28 instead of doing five different steps,
30:30 so we can condense that iteration cycle down to one step.
30:33 And Rockbot as an interface already has a skill called learn from demonstration,
30:38 where it's telling Maya in this case of being able to actually turn
30:43 this demonstration into a reusable skill.
30:46 And so this is something where, again,
30:49 now if I am working on onboarding Maya to do specific flow in my company,
30:55 being able to actually give it the skills and workflows and pass this off as
30:59 a template.
31:00 All my other teammates across the board knows exactly how to operate on
31:04 this specific job.
31:06 And so it looks like, got it.
31:08 When we do the sticker vote, I'll publish the form, like a QR for the audience.
31:12 This looks solid so far. You created a blank Google form.
31:16 So, awesome. It finished watching and actually save it as a skill for now,
31:21 which is great.
31:22 What I want to move into now is actually having new bots now work in
31:26 sequential order and talk into each other.
31:29 So we've trained Maya now to be able to build its Google form once we have
31:33 the stickers.
31:33 The next thing here is that we actually need to know what sticker to
31:38 actually do. And so what we can say is,
31:42 can you message I would do at Dex that
31:49 you are able to create Google forms and that there is an
31:55 audience that will vote sticker?
32:01 Work with Martin to create the stickers and don't do
32:06 anything until you speak with him.
32:12 So now this is more where we see,
32:14 again, these team of bots of not only are they able to actually handle a
32:18 slice of your role,
32:20 but now able to also delegate that work onwards once it is able to complete it.
32:24 So just the backtrack here before moving on to Dex is
32:28 that we realized Maya actually saved a skill called Create
32:32 and published Google form.
32:33 So now intelligently,
32:35 it will know to apply this next time if my request in prompt is applicable
32:40 to this. Elsewise, it won't bring it in if I have many other skills as well.
32:47 So cool, if you notice, Maya message decks.
32:50 And the cool thing about Rockbot is that we actually have transparency into
32:53 what they're doing.
32:54 So Maya says, hey, I'm Maya, this is what I do.
32:58 Please work with Martin here and actually creating the sticker options.
33:04 So now I have. actually go into decks
33:08 or it looks like iMessage the wrong decks here so I actually say
33:15 sorry message the wrong decks message this one moving
33:21 forward post
33:28 that one out and so now decks again all of these bots across the board I
33:33 really want to showcase how we can start from scratch and get that working
33:36 so now we can see that I got a message from Maya we can move
33:40 and do it exact same thing and so this is now my is ready to build a google
33:45 form what should this stick your idea be what is great is
33:49 that this is where we start thinking about the connectors
33:51 that we can give Grok Bot whether in this case it will be Figma MCP
33:54 which we can look in the marketplace but for this example i have a specific uh
34:03 Figma guidelines that we have here at Cursor and space x AI for
34:07 when we want to create Grok Bot apparel stickers,
34:11 there is this specific icons and assets to use so that our deck designer is
34:16 able to actually use it and then create really cool,
34:18 hopefully looking stickers there for us.
34:20 So I'm just going to paste in this specific prompt,
34:24 create eight Grok Bot stickers following these guidelines from Figma,
34:28 and then just having it just tell back to us what exactly it's thinking
34:32 before we even move it along in the process there.
34:36 And so now if we really think about being able to use computer use for each one,
34:42 going to connect apps and Figma is that we have these out-of-the-box
34:48 connectors that are bidirectional.
34:50 So not only we're able to actually pull context in from these sources,
34:54 but we're also able to update those downstream objects as well.
34:57 But what I want to know is that Figma and their team has configured a set
35:00 of tools,
35:01 which is just functionalities and things that AI is able to do with Figma.
35:05 And so out of the box,
35:07 there's a lot of good things like getting a screenshot, generating design.
35:11 But as you know, Figma is a very complex platform.
35:14 And so there are some things we might be limited to doing
35:17 if we solely just use this connector or MCP,
35:21 which is why computer use exactly important and why it's awesome for DECS
35:26 to be able to actually go and Figma view it exactly how we view it.
35:30 And again, we can even teach a task to DECS on how to
35:35 create slides in this case also how to create stickers so it'll say i'll
35:40 tell you what i see before you make any stickers assuming it's different
35:44 things i'm just going to copy the second prompt here once
35:47 it actually does tell me what the stickers are going to do
35:57 and so now it mentioned got closer shots of those brain assets colors
36:00 and shapes this guideline is pretty intense in the terms of there's
36:05 different tabs different things you can actually use um for pulling in
36:09 that context from Figma
36:11 And now it'll be like,
36:12 this is what we'll summarize for what actually matters for these stickers.
36:17 So to Dex's point of view,
36:19 does not matter what Maya is talking about in terms of making Google forums.
36:23 This is where we view, like when you want to create a bot,
36:26 this is more or less just being very in tune with what that slice of the pie is.
36:31 And so it mentioned that here's what the Figma Links opens up into.
36:35 You have characters, colors, stickers.
36:38 And so even giving us screenshots because it's able to.
36:41 to open up in Figma's computer, what exactly we're working on.
36:45 So I'll say, awesome, I want eight new stickers.
36:49 Pull up the new Figma file for these stickers once
36:56 you create it on your computer so we can watch.
37:06 Awesome.
37:07 So I'll take a pause here and we can wait for Dex to actually open its
37:12 computer, open Figma.
37:14 Are there any questions here?
37:16 specifically Cliff that you've seen have been surfaced?
37:20 So many questions.
37:23 I think a couple of common patterns are how to think about knowing
37:27 when you should use like computer use or like the in, you know,
37:31 the actual computer browser yourself to teach it a task
37:33 or something versus letting it go on its own or trying to find a connector.
37:38 And another common question is around this idea of context windows
37:42 and whether or not that applies to Grok Bot.
37:45 since I think folks are familiar with that from working with other
37:48 coding agents,
37:50 I'll let you answer that one.
37:52 Awesome. So for the first question, we can tackle there.
37:56 It sounds like it's more or less of how to use computer use,
37:58 or when using computer use, whether to do teach a skill or let it roam free
38:02 and actually do the job.
38:04 Is that the first question there?
38:05 Yeah, that general realm of like how to think about that. Awesome.
38:09 So as X pulls up, I think I'm on computer and we can watch it in real time.
38:13 The way that I view this is more so guiding the Grok Bot behavior in a
38:19 way that makes sense to us.
38:20 So we could let our Grok Bot run free.
38:23 And it is typically pretty well actually designing and going end to end to
38:28 end on creating those stickers.
38:30 And so in this case,
38:33 like if we want to actually save in tokenomics and the API calls behind
38:37 the scenes, we should be able to actually create a skill, which now.
38:43 now tells it step A, step B, step C,
38:46 and it's able to get to your step A to point A to point B in a more
38:49 efficient manner.
38:51 So it's more about like actually reducing the iteration cycles.
38:54 This is where you would use the teacher task with computer.
38:57 Now in terms of the plugins,
39:00 if you're able to do a lot of the flows with pulling plugins,
39:03 that is also very simple to do.
39:05 When I think about computer use,
39:06 it's just more of things that you are unable to do with plugins.
39:10 So my example of Figma, having those specific tool calls can be very limited.
39:16 And so that's when you want computer use to actually be able to use Figma
39:19 for the platform that it was built to use.
39:22 So when the connector was created with Figma,
39:24 it was more or less for the points of AI to only be able to pull data from
39:30 that source.
39:31 Now with a computer use,
39:32 it's now a whole new ballgame as to what we can actually do.
39:37 Awesome.
39:38 And so what we'll notice here is that Dex was able to actually showcase
39:44 building stickers we can even pull up directly the Figma
39:51 here looks like it erred out and so at this point is i would go back
39:57 and forth with decks to can iterate on it this is a one shot trying to
40:03 actually onboard decks into how i want to design Grok Bot stickers this is
40:07 where we see actually taking the time of if i'm a designer where i want decks
40:11 to not only handle stickers but also assets to be able to teach it in
40:15 plain text of here are inspirations here's different things of actually how
40:19 you should go about building these stickers out and so if on the first
40:23 pass i'm not extremely satisfied with it which i'd be extremely happy
40:27 if it was able to one shot this is where i now course correct it
40:31 and teach it exactly what it should do and so behind the scenes i ran decks
40:36 a couple of times it gave me different outcomes just because i did not
40:39 actually to cliff's question create a skill to build the stickers in a
40:44 way that's routine and consistent.
40:47 And so this is where I actually have different sticker options here
40:51 that we can actually showcase.
40:54 And so this is some stickers that a previous Dex made and
40:58 that I actually want it to use for the sticker vote.
41:01 So what I'll tell Dex is awesome,
41:05 notes the actual Pygma screen on your computer.
41:10 I want you to send
41:14 Maya
41:18 that screenshot
41:20 so she can create a Google form and have the audience vote
41:27 in real time.
41:32 And so now this is more so not closing the loop.
41:35 So Maya is talking to Dex.
41:37 Dex is now talking back to Maya.
41:39 And if we now go because Dex has its own computer,
41:43 able to actually create a screenshot now of this Figma on the computer.
41:49 And so all of these as like a conceptual context of how we actually give
41:54 these bots computer, for each Grok Bot user,
41:57 so you're a user, we provision an entire VM or computer for your use,
42:03 and then each of these different Grok Bots have different desktops within them.
42:07 So if you notice, Dex is looking at Figma on that screen.
42:12 Maya, though, is using Google forms on a different screen.
42:16 And so it's as simple as having shared credential access.
42:19 So once actually log in to Google on this computer from Maya side,
42:25 Dex is also able to do that as well.
42:27 And so now what we'll see is based on the skill
42:31 that Maya learned from me teaching a task,
42:36 Maya is hopefully now going to create Google form,
42:39 post that screenshot, and find a way to do so.
42:42 So it saved the photo into its computer and we'll now post
42:45 that as an actual photo for y'all to vote on here shortly.
42:51 So let's see, Maya's not going to be working.
42:55 We can leave her computer up here.
42:57 We can see that now a new Google form was created.
43:01 We can even just go into full screen here.
43:04 And we can see what Maya will now do in terms of creating the title description
43:12 notice that it already took a picture of that Figma,
43:15 and so it knew the different options here.
43:18 It looks like it's already ready to publish this form.
43:22 And so what I want to note is,
43:26 can you put the screenshots of the different sticker
43:31 options in that question,
43:40 just so that when you actually do scan the QR code,
43:43 it is very easy to do that there?
43:51 Awesome. So we'll let that work there.
43:53 Cliff, is there any other questions that have surfaced that you've seen?
43:58 Yeah, a couple. I'll just answer for everyone is yes.
44:02 All of your bots are using one computer.
44:05 It's like one cloud VM,
44:06 but each bot is going to have, how would you describe it,
44:09 or like their own screen or their own desktop, if that makes sense,
44:11 which is part of how they can parallelize the tasks.
44:14 And again, in Grok Bot, you're not supposed to,
44:17 or you don't need to worry about things like the idea of a context window.
44:20 Think of the bots as a long-living teammate that are going to manage
44:24 many, many agents on your behalf, each with their own context window.
44:29 And so it's going to do all the coordination of, like, you know,
44:31 creating sub-agent threads, managing their context,
44:34 managing their memories and tools on your behalf.
44:38 Other than that, nothing big, I don't think, Martin.
44:40 Yeah, exactly. Some questions about the VMs.
44:44 Many, actually, many questions about, like, where are the VMs hosted?
44:46 What are the specs? Can we change the specs?
44:48 But probably beyond the 101, unless you have any.
44:50 high-level thoughts you want to share.
44:53 Yeah, I think in terms of the context memory,
44:55 the main gist of Grok Bot is that we want to keep it as simple as possible
44:59 to use. We realize that even if you have an AI tool that is powerful,
45:04 that is able to have its own computer, that is able to work autonomously,
45:08 it means nothing if it's not as simple as spinning up a new bot
45:11 and talking to it in plain language.
45:13 And for us to be able to handle that memory and context behind the scenes,
45:16 that is something that you don't have a threshold here of like,
45:20 you've talked to your bot too much today,
45:23 it's going to summarize and lose its threaded memory there.
45:26 So that is less of what's actually going to happen.
45:29 And yeah, we handle all the context memory behind the scenes.
45:39 So now it's moving that image.
45:41 As you notice, I said that that is the wrong one.
45:43 Use this specific one instead.
45:46 And so now this is where it will actually
45:49 create the question of what sticker we should print.
45:52 And then specifically that's Korean shot in that Figma there.
45:55 So now it'll copy their response under link,
45:58 hopefully now create the QR code that the audience will then be able to
46:02 actually do in the last remaining time here.
46:05 And so while this is working is I can tell Penny and we use Grok Voice,
46:14 hey Penny, just to note, so moving forward,
46:18 we have an audience watching us today doing a 101 demonstration of Grok Bot.
46:23 What we've had so far is we've had decks actually create Grok Bot stickers.
46:27 We've had Maya create a Google form so that the audience is able to ask.
46:31 actually vote on what sticker we should print.
46:34 Note that I'll give you that information when we get there.
46:38 I just want you first to look for vendors
46:42 that in the New York City region that would be able to actually print
46:45 stickers for us.
46:46 And look online, find that, and then come back to me. Are there any questions?
46:51 No questions. NYC sticker printers online coming back to you.
46:58 Awesome, thank you.
47:00 Glad to help.
47:01 Want me to flag anything specific about the stickers while I dig,
47:05 size, quantity, rush, turn around. Nope, we can end the call.
47:10 Sounds good, Martin. Talk soon.
47:13 So we can see that once we have the voice chat,
47:15 it's able to actually close off the voice because it knows
47:18 that I'm moving on to a new task.
47:21 We even have the dictation here and that summary.
47:24 But notice now that the sticker vote forum is now live.
47:28 What I would ask the audience to do is if this is on your computer,
47:32 And if you're on your phone,
47:34 be able to actually take a picture of this QR code
47:38 and vote for the actual sticker that you like here.
47:43 And so just to note and highlight is that if you're on a Safari,
47:49 sometimes it might not load the image. I would use Incognito or Chrome tab,
47:54 but really be able to actually vote now on the sticker and then we can wait
47:58 and get the responses in here. And
48:18 give me a screenshot the six stickers and the QR code in one image.
48:23 So it's easier to vote.
48:31 Are you seeing any messages here, Cliff, that they're unable to vote?
48:34 Or actually, it looks like we are getting responses.
48:37 We have 415.
48:41 Go team.
48:44 Cliff, not that you will influence a decision here,
48:46 but what sticker is your favorite so far? Which of these?
48:51 Yeah. I like the bottom middle one. It looks like two friends.
48:55 Friends. Your friends?
48:58 That's like my index in this case.
49:00 So here, notice that it just gave me a screenshot so it's easier now to vote
49:04 and view the stickers to your left.
49:06 Again, literally whatever Rockbot you wanted to do, it has some computers,
49:10 it's able to do that in this regards.
49:12 And so now we'll just, for the sake of time,
49:15 a couple of minutes left, there's actually close out this thing and say,
49:21 tell Penny and you work,
49:25 can you get the sticker winner?
49:27 from Dex and
49:31 then have Maya give you the actual image or
49:38 that winning sticker and pull all of the different
49:43 vendors we can use here and I will pick.
49:52 And so it should be able to find that Maya is referring to.
49:58 I can just say specifically that Maya is this one here.
50:02 And so now when we think about the overall trajectory of what we were
50:06 working on,
50:07 of being able to now onboard Dex, a designer to build the Grok Bot stickers.
50:11 And again,
50:12 that was a one shot of me, would actually work meticulously with Dex to do so.
50:17 Onboarding Maya to be able to use the computer use, to teach the tasks,
50:21 to train, to use skills there so that it is able to do this sequentially on
50:26 its own if there is a new product that we want to operate there.
50:30 Notice that even got a message from Dex,
50:32 and now it should check the form responses for the winning sticker.
50:35 and then to even to penny.
50:38 So across the board, they all share the same quote-unquote computer,
50:42 but the different desktops.
50:44 So we wanted to keep it extremely easy that if you sign in once on one
50:47 computer or one desktop, rather,
50:50 it's synced across all of your different bots along the scenes.
50:53 So it's now looking into the registration.
50:58 We can see that looking at you did actually win.
51:02 So congrats there, Cliff, on your choice.
51:07 Hope I didn't skew the outcome.
51:10 Fortunately, it was already winning, so good there.
51:14 But
51:16 it looks like I used the wrong names.
51:18 So notice I said I can get the sticker winner from Dex,
51:21 but Maya actually knew the winner.
51:23 And so we can see that because it talks back and forth,
51:27 we actually were able to see in real time of the fact
51:30 that they actually figured it out and that it actually was from Maya
51:34 that was the actual bot that should pull that information from.
51:37 So we now have our own A-Sing team that's running and working together.
51:41 Again, we can see the conversations of Dex saying there's actually a
51:45 winning count vote looking at you one.
51:48 And so here is Penny actually doing a preview of the sticker.
51:53 I will say, awesome, I can print a specific print leaf Midtown.
51:59 Open this up on your computer and upload a sticker, J-Pagic.
52:08 So now we're just going to wrap it.
52:09 up here in terms of procurement and sourcing,
52:12 Penny will now open in its computer,
52:16 specifically what this vendor would be pulling the actual Figma asset into
52:21 his computer and then uploading it to the vendor of its choice here.
52:26 And so I do want to be cognizant of the remaining four minutes here.
52:29 I just want to kind of recap everything we've talked about where Grok Bot is
52:33 the first ever teammate where you're able to onboard it in natural language
52:37 and to give it a role,
52:39 to give it a job and as we move up this maturity curve,
52:42 able to actually have potentially a now chief of staff.
52:47 So I can, what I can do is promote Bobo to not only be working on personal
52:52 life stuff, but now actually, hey, you have a team.
52:56 This is where you're going to actually be working end to end together
52:59 and you are now their manager.
53:02 So I put that in the Rockbot sticker team section.
53:05 Doesn't actually create a sticker just yet.
53:09 or it does actually create that routine between them just yet
53:12 but what i can do is um
53:15 i get out of this create a group chat and this is where i put penny i'll put
53:21 dex i put maya and i put bobo and i'll say bobo you have
53:27 been promoted to chief of staff
53:33 dex penny and maya can you say in here what you've
53:40 done in regarding to Grok Bot Snickers so
53:45 that Bobo is pot up speed.
53:50 And so the interesting part is you'll realize they'll all now respond to
53:53 each other in a group chat fashion.
53:56 And this is where now, instead of me having to manually talk to Dex,
54:01 manually talk to Penny, I'm able now to actually only talk to Bobo,
54:05 which is the chief of staff,
54:07 and that was in that SDR's Grok Bot Galaxy case, the Josh Allen chief of staff.
54:11 Same thing.
54:13 of now being able to actually speak to one leader that knows exactly how
54:19 this team should operate.
54:20 And that only happened because I actually took the due diligence
54:23 to train each bot independently and now give them the actual skills
54:27 and routines they need to go end to end here.
54:31 And so we'll notice that this is more in the 201.
54:33 Joseph will really showcase the different types of things we can do across
54:37 the board with team delegations and passing back and forth.
54:41 But this is with a fundamental two minutes here, the Grok Bot interface,
54:46 and hopefully there is something here from computer use to connectors on
54:51 the bottom left hand side that you can take away in your jobs
54:54 or even personal life there.
54:57 Awesome.
55:00 So just a couple of summaries just to wrap up is
55:03 that bots can access every single tool even once without MCP.
55:08 So this is the ability to create that Google form,
55:11 to use Figma the way that it was intended to be built for users.
55:16 Bots have memory, and we handle all of that memory for you,
55:19 so you can talk to the same bot until the end of time.
55:22 Well, actually, wouldn't quote me there,
55:23 but you can talk to the same bot for an extreme long amount of time there
55:26 to actually learn how you operate,
55:28 so you can trust it to do actually meaningful work end-to-end without
55:32 your permission at every step of the way.
55:35 And lastly,
55:37 when we are doing shell work,
55:38 we're doing any specific work that we wanted to get approval, action from us.
55:43 We are able to actually go with the last thing here.
55:45 I want to showcase settings.
55:48 And we have specific rules where it's that when Grok Bot wants to say,
55:54 create a Google form,
55:56 it should ask first before actually creating and going doing that.
56:00 So this could be sending emails.
56:02 This could be responding to someone on Slack and you don't want them to
56:05 know that AI responded. You would have all of your rules here built in.
56:12 Awesome. So, I want to leave some time for Q&A as no,
56:15 that didn't quite match out there.
56:17 But hopefully, I know Cliff was probably locked in on that Q&A section.
56:21 Thank you so much, Cliff, for handling all of that.
56:24 This is also, I believe, the number one workshop in terms of registrants so far.
56:30 So huge kudos to Cliff there and managing all of that there.
56:34 And thank you all for joining.
56:36 Thanks, Martin. Thanks, everyone for coming.
English transcript generated automatically from the audio and may contain errors. Bulgarian subtitles: BalkanDub. Download subtitles (.vtt)