Grok Bot for Enterprise Teams: Put AI teammates to work across every department

0:02Um, cool. Well, thank you all for for coming. Um, Anita, do you want to kick us off? Yeah. Um, so excited to have everyone with us today. Amanata, I'm on the AI deployment team here at SpaceX SI. And we also have Santi with us from the field engineering team. Today we're going to be talking about Grockpot for enterprise and specifically how do we create these really great AI teammates that can work across multiple departments. So going to give it over to Santi here to take it away. Thank you so much. Uh yes, hi everyone. My name is Santi Garza. I'm a senior field engineer here, which just means that kind of the

0:35technical side of a lot of our products and help uh people like you and companies uh adopt the products uh to the best of their their ability. So a lot changing right now uh in this world of AI. And so my job is really to make sure that our customers are experts on how to build with AI. So um today we'll be uh covering Grockbot and how it can be used to increase efficiency across different departments um within enterprise teams but this also does apply to uh startups or even uh oneperson teams. Uh so I think you'll get a lot out of this regardless of of

1:15who you are and where you're coming from. Um, and I'll be uh covering various use cases uh throughout to give you an idea of what's possible, but um happy to uh brainstorm a little with you at the end uh during Q&A. Just a reminder to drop your questions in the Q&A chat. A lot of them we'll answer uh in the chat, some of them will answer live, and we'll try to get to as many of

1:38them as we can. Um so, just uh real quick some housekeeping. Um, if you signed up for this online, you should be receiving a recording shortly thereafter with the link if you want to rewatch. But, um, we'll kick it off with kind of like the, uh, the very recent but sharp maturity curve, uh, in the world of AI. Just a few years ago, we were all dealing primarily with chat. And so, um, that, you know, that kind of felt like you were calling, uh, uh, an increasingly intelligent friend and, uh, you would be able to send out a question and you'd get an answer. It felt a

2:18little bit like the evolution from maybe Google search. Um, but then last year, um, our agents started actually being able to edit files on our machine and so like build websites, build applications, and that was very, very exciting. Um, what Grockbot popularized though is this new idea of real teammates. So, having agents that have an identity and they have a job and they they know what kind of work they're assigned to do and you can really count on them to work even while you're away. So, you don't really have to babysit. You don't have to play ping pong with your agents. And so, that's what we'll be talking about

2:57today. I won't spend too much time here because this is primarily around Grockbot, but I just kind of want to give a lay of the land here. Here at SpaceX SI, we have sort of four product categories. Grock is the main model. So, we had a big leap in intelligence the last two versions. So, Grock 4.6, Grock 4.7. Um, and uh at

3:21at a great price uh for the frontier. So, Grockbot is uh where we'll be spending most of our time. This is an application where you can actually uh hand off tasks and start delegating um routines to your teammates. Um the third category here is cursor. Um cursor was recently acquired by SpaceX and Cursor

3:43is really uh an agentic coding platform. So for engineering teams but also for people who are new to uh to development. It's a it's a great tool. It's very robust and um and you know the philosophy there is that you probably shouldn't be vibe coding your way to production. So uh cursor is a a leader in that space. And lastly, Grock API. So Grock API lets you um actually put the the uh the models inside of your products so that you can power some features and and products with the Grock API. So um I'll just really quickly breeze through. Um, the cool thing about Grock is that it has a great integration

4:21with X as well. So, it's a truth seeeking model, but also does great sentiment analysis. So, you can see real time what people are saying and how people are feeling, maybe about your product, maybe about your industry. Um, Grockbot is uh this is this is the Grockbot interface. Some of you have seen it. Um, we're going to be spending most of our time here today. And so, on the left, you'll see that each bot actually has a name and and has a a kind of a unique job description. Um and then you have the conversation in the middle which um is persistent uh so it learns

4:53with you over time and then we'll dig into some of the features uh during the demo. Cursor uh just for any engineers in the room or QA people or designers who are working a little bit closer to an application or a website. Uh cursor lets you build in parallel. So cloud agents are one of our most um incredible features. I use them every single day and uh they allow you to build uh on the cloud also on their own computer similar to Crockbot. And so we'll be seeing a little bit about that today, but you can uh kick off work from your phone, from

5:28your laptop, from Slack, from Linear. And lastly, um the Gro API. So, uh we'll be uh running a demo on this today. Um but uh Grock lets you uh pick up the phone and make a call and and it'll answer. So it's a it's actually the number one model at the time of recording for um for voice for speech speechtoech voice but also our Gro imagine is is incredible for image and video generation really fast um and really cost efficient. So without

6:02further ado let's hop into the demo. We're going to be working today in uh with a a madeup company uh called Flyo Airlines. And uh this is a this is a uh a madeup um airline. It lets you book through here um you know everything you'd expect. So um let's uh let's talk about Grockbot. So I'm just going to pull up the visualizer and I if you want to just uh are there any questions yet? I I probably not, but um yeah, I Yeah, I mean I think one question that's coming up often is like what does voice mode inside look like right now? Can Crockpot chat with you

6:47versus you just chatting with Crockpot? Yeah. Um do you want to do you want to take that one? Yeah. Um happy to. So yes, Crockbot can interact with you. Grockpot can leave you voice notes right now. And you can actually see how to switch uh the voice for Grockpot as well. So definitely something we recommend trying uh while Santi gets his systems up if you guys

7:11wanted to take a shot right now. So this is a um this is a visualizer. This is not the the actual product, but it's a it's a a really helpful tool that we use to try to explain this new shape because it is unfamiliar. Um, the way to think about Grockbot is if you could hire a teammate, if you could hire someone to work uh for you, under you, who would be

7:35the first person that you would hire? And for a lot of people, that's it's you the answer is usually an assistant. And you know, there's a lot happening in the world right now. I'm sure you're just like me. My plate is overflowing with emails and Slack messages and product updates. and then the industry is moving so quickly and so having an assistant to try to manage all that and manage a schedule is a really compelling first uh first use case for Grockbot. But once you have the assistant, which is super helpful, who else would you hire? And this is going to um depend largely on your role, on your team, your

8:14department. But for example, uh if you're an engineer or you work with a product and there are constant uh updates shipping, maybe you would hire uh someone to help with releases. Maybe you would hire someone to help with QA and testing. If you are in sales, maybe you would hire someone to uh keep your prospects uh warm and to draft your emails and to check in on your clients. uh maybe even join your calls or follow up on transcripts or promises that you made to your clients. Um if you're in IT, uh support tickets, right? I mean, there's a lot of help that uh you'd probably

8:57hire in. And so, um that's that's how we urge people to think about Grockbot is where do you need help? Because as you'll see today, you can actually help uh you can you can train in um a new bot that can actually take a lot off your plate. sometimes even 100% of a task off your plate. So, um we're going to start

9:19here with this visualizer. And uh what this visualizer does is I'm just going to move these over to the right. Um like I said, this is just a an internal tool, but I can pull down let me um is this helpful if I make it a little bit bigger? So these workflows here are different workflows um different routine tasks that get handled by uh one or more of these bots. So um we're actually going to look at a few of them right now. So I'm going to start with

9:53the support called tofix this one here. Okay. So what this uh workflow does is it uh it handles when someone calls in needing support. It handles uh the the call and it also if it's an issue with the product or with this with the website for example, it'll actually try to fix the issue. So uh there are three bots involved here. So let's start with the first one. So the concurge. So the concurge will actually answer the support line. Now in this case it'll actually handle it once the transcript once the call ends it'll receive the transcript but in some cases you can have the bot actually listen live. So um

10:38the records bot is going to manage the past history and maybe any promises that we had previously made to a caller. And the third one is the QA bot stands for quality assurance. And what it what it's going to do is if this is like a bug or an issue with the website, for example, it'll check on its own to see if it can reproduce that issue in our system. So, we're going to go through the workflow here so you can kind of get a um an idea of, you know, how to start thinking about these systems. All right. So, we'll start here with the concurge um

11:13bot. And so, the first thing it's going to do, you'll see on the right hand side, is this uh text too small? Can you guys see this? No. Looking good. Looking good. Okay. So, on the right hand side, you'll see the steps. So, this is step one of the workflow. So, open the new case in the help desk and read the whole transcript. Note who called, what they were trying to do, and what went wrong. Now, if this is a returning caller, it's going to kick it up to the records bot. And what the records bot is going to do is it's going to look at the caller's past cases. What

11:46did we promise them? how do they like to be contacted, etc. Um, this gets skipped if it's a first-time caller. And so it'll pull in that history and write that into the case notes. Then it'll check against our policy. So if it's a support call that, you know, needs to be maybe checked against our policy, that's what happens here. And then it'll make a decision. So if the policy answers the question, it's going to just draft a

12:12reply to the customer. if this is actually a bug or or it thinks that it's a bug uh in our system, it'll hand it up to the QA bot that knows how to test if something is broken. So, it has its own computer, it has its own environment, and I'll show you this in a little bit where it can actually see if it can reproduce that issue on its own. And if something is serious, so if there's um a risk or someone was charged wrongly or you know someone's in danger, something serious is happening, it'll escalate it up immediately to a team. Um and and you

12:51can set this up however you want. I mean, this could be uh something that gets escalated to your team on Slack. It could be um a phone call. Uh it could be whatever you set it up to be. Once it has its drafted reply, it actually stops working autonomously. And this is because it was set up this way where it'll draft a reply and kick it up to you and wait for you to send it. Um, this is helpful because sometimes we want to just make sure that um what we're sending out to our customer is exactly what we want the customer to receive. Um, but there are some

13:26workflows where if it's a low-risk situation, you can actually trust the bots to manage the communication on their own. Once uh you send the reply, the concurge bot will close the case and tell the records uh bot what happened. And this is really really helpful for the records bot because the um records bot will then know, you know, for similar sounding cases. It'll be able to reach back and see how it was handled in the past and perhaps expedite the process on the next

13:55run. Um we're going to look at two more just to get your ideas going. So, um I'm just going to go back to the visualizer and we're going to choose um this one here, low stock to order. So, um this workflow will check for inventory that's low in stock and it'll order it at the right time. Um so, for anyone who's not working in software and act or you know, even if they are uh for offices and things like that, um logistics, um I'll I'll show you kind of how this goes. So, uh, the plannerbot will watch for the stock levels and flag when it's time to reorder. The

14:37storekeeper, uh, monitors incoming parts of, um, on the ops platform, so it's hooked up to the ops platform and it knows when there are parts coming in already. And then you have the purchaser, which will draft, um, the supplier orders for for approval. Um, so I'll show you one more here. We're going to do Where is it? Here it is. So, this prospecting plan here. Okay. So, this is a like a higher touch uh workflow. So, um it actually does most of the heavy lifting, but it respects the the fact that this part of the business is like a human to human world. And so, um so this

15:22is a a sales workflow. So, um, what this workflow does is it it builds a prospecting sheet from your Salesforce account. So, this prospector bot will actually have been signed into Salesforce and can do this um with uh I'll show you how to set this up uh in in a second here, but it can actually uh work within Salesforce. Um, and so it'll uh find the right companies and people to contact, who already uses your product, and then fresh reasons to reach out. And um and it stops at every uh key moment along the way to check if the accounts are okay and you want to

15:59proceed to check the drafts uh the outreach drafts and to make sure that the samples are okay. So you can set this up however you want. Um but here you're probably cutting you know 90 to 95% of the time of this particular workflow without cutting out um you know your authority and and kind of your

16:17influence on the matter. So, and of course there are just endless routines that you could imagine Grockbot picking off. Um, internally we run thousands of routines and uh we each have our own personal bots, but we also have team bots that run these systems across like an entire department. And so really what it feels like to us is it feels like we're we've become sort of managers. Um it it feels a bit like a promotion to be honest of having to of being able to bring people in under you to start to delegate work away and start to focus more on the systems of work uh

16:56without losing the um you know the focus and and kind of those important moments where we want to be uh active. So all right let's look at a a marketing case. Um I'm going to open up uh Grockbot here. And I might just make it a little bit bigger. Okay. So, this is uh this is Grockbot and um we're going to run through a few

17:26different cases. So, we'll start with a a marketing case. Um so, the world is moving very very quickly right now. Um, and it's becoming increasingly difficult to stay updated on the competitive landscape in your industry. And so, um, you know, this is something that we have to do constantly as well. And so, let let's see if we can get some bots on that. So, creating a bot is easy. You can just go up here to the plus sign and you can hit create a

17:58bot. And uh because we are um let's name this bot um competitive um Carl. And um it's Halloween, so I'm going to choose this fun little pumpkin one. Um I'll create the bot. And so the first thing that's going to happen is the bot is going to um based off of its name suggest you know a first step. So um it's saying what do you want me to help with? Track competitors pricings pricing and fairs watch competitor news and launches and build uh competitive briefs for deals. So it's giving us some some ideas. I'm actually

18:44going to give it a custom brief here. So, I'll say you're Flyo Air's competitive intel analyst. Your job is to watch competitor fairs and route announcements for our top routes and tell me what changed and what we should do about it.

19:08Now, uh, typically I would spend a little bit more time and the way I like to do it is I like to actually write a job description. Um, I find that that helps me structure the work a little bit. Um, but if at any time you want to evolve the work, you can actually just um, uh, prompt it and it'll it'll learn

19:29to evolve into other spaces of the role. So for example, also track changes to competitors loyalty programs going forward. And so this is really an iterative process uh similar to bringing like a an actual teammate onto your team. So um okay, so it says got it. It's going to uh watch competitor affairs and then it's going to set up a a daily check once once it has them. and it's it's got this follow-up instruction. So eventually uh what might happen is you might work back and forth with this bot and it might accumulate quite a lot of context, quite a lot of

20:09responsibilities. And so at any time you can actually I'm just going to show you a model that we've started to embrace internally. You can actually promote the bot. So I'm going to say I'm promoting you to director of competitor intelligence. I want you to build three new bots that report to you. The first bot should check competitors prices across similar flights. The second bot should scrape our competitors social media accounts daily and detect what's landing well with their customers. And the third bot should check their websites and promotion

20:44emails for important announcements. Now, I'm just trying to get your ideas going here. Um this is uh these are great use cases. Um the for example checking websites constantly to see what competitors are doing. That is a very tedious task but gathering that intelligence is super helpful to you and the business and logging it somewhere where you can actually tap in uh maybe using Grockbot to come up with strategies or counterpositioning whatever you might want to do there. Uh the other cool thing about this is that um you noticed uh pro I mentioned promotion emails. So um Grockbot actually can now have its own email address. Um I encourage you to secure

21:30your emails soon. Um but uh it it can have its own email address and so it can subscribe to those kinds of things and uh make a lot of sense of out of all the noise. So okay, it's thanking me for the promotion. its title changed up here automatically to director of competit competitor intelligence. And you'll notice that it created um three new uh bots here. So these three bots are going to report to this bot. So we're starting to build our little organization under us. And um and so uh this is helpful because uh as you start to build out more and more bots uh on your team, the

22:13same thing that would happen with like a real team growing uh those growing pains are going to happen here too naturally. So actually when I started using Grockbot when we had it internally it was so powerful. I could take on so much more work, but at the same time I was like, "Oh my god, we have bots to manage." And so I actually uh went and read some some um uh you know, psychology, some kind of like organizational theory, and I tried to figure out like how do the world's best CEOs um or managers uh direct and manage larger teams? And uh one thing that was

22:53really really helpful is to make sure that you're only interfacing with between one and seven teammates uh reporting to you I mean and so uh you can funnel the communications up and down. So, for example, in this case, I can speak directly to competitive Carl, which will then propagate and decide and direct and manage the bots beneath and we'll check in on the bots and make sure that the bots are doing things, you know, kind of strategically the way we want to. Um but also it means that um you can actually instruct the bots um to to organize their um their reports and their communications in a way that is is

23:36more efficient. So for example um asking them to show proof of their work, asking them for links to their sources, things like that. That'll propagate up to the manager which will then have that context for you know managing the team going forward. And that way it'll only bubble all the way up to you if it's something that needs your attention or you need to be aware of or needs your decision. So just a quick aside on how to manage these bots. All right. So let's go over to this social uh bot. So um okay. Sometimes you'll see these little cards here. Uh so what it's asking me is it's asking me if it can

24:15save a routine. So, as as you remember here, um this one's going to check uh against some of our competitor airlines, and it's going to, you know, on socials, um including X and and LinkedIn, etc. And so, what it's asking here, let's let's pop it open and see. It wants to be able to review the last day of public

24:35social media posts from the competitors. And it wants to write a daily routine. So, I'm going to hit allow on this. And so you'll see up here in the top right that we have a new routine for this bot. By the way, this routine is something you can edit anytime by working with the bot and saying, "Hey, actually, I only want you to do that once a week on Mondays, for example, uh, and look back at the full week if it's a less urgent thing." And um, and so it'll stop and ask you for um, uh, permission on the things that uh, kind of need your

25:09approval. Um, so they won't go they won't go rogue on you. Uh, once it's all set up, it'll run regularly on its own and you don't really I mean you can rest rest easy knowing that that job is getting done. Now um sometimes it's easier to uh to teach a task than it is to explain a task. So, um, what I can do here is I can actually record myself, uh, doing a task on the bot's computer. I mentioned briefly that each, uh, bot has its own computer. This is, um, this is a big deal. It sounds like a small thing, but it's actually a really big thing. So,

25:50let me see if I can, uh, pull one up here so you can see. So, we're going to go to the fair scout here. And um you'll notice here on the right hand side, this one's asleep. Uh the it's not currently using its computer, but I can open up its desktop and I can see that, you know, I can take a peek into it like a window. Um every bot has its own desktop. They actually per user account, they actually share one uh computer uh a virtual machine. um which is really really helpful because that way you don't have to continue to explain to each bot you know um you know

26:30certain they can all have references to certain files um and also like if you use a credential uh frequently um and you want the bot to be able to log in there you don't have to log every single bot in uh but at the same time they each have their own desktop so they can parallelize across different tasks so um

26:49I'm going to teach it a task here. I'll uh down here in the plus button, there's this option to teach a task. And so I'll hit that. And so uh I can hit start recording. And so we're going to pop open uh Chrome here. And I'm going to say flights.google.com.

27:15And it's recording up here in the top right. So, I can say San Francisco to Vegas, for example. And we'll do a uh let's do a one-way. And what I'm doing here is I'm going to check to see which uh one week out which airline has the cheapest fair and what is the fair. So, I'll hit search. Okay, so it's Frontier again. Uh, and it's $84 is the cheapest uh, fair one week out. And so now I'll switch over to, let's see if this will pop up. Fly

27:54lowair.com. So this is our uh, sales.flyair.com. It's an internal tool here. Um, so okay. So I'll I'll log in here and I'll say uh, admin at fly low. This is our fake uh our fake account here. And let me see if I can Oh boy, this is gonna be tough. You know what I think what I'm going to do? I'll show you actually uh let me quit that real quick and I'm going to show you actually a really helpful tool here. So I'll say uh log into sales.flyair.com

28:38on your machine. So what's cool about this is it's actually going to ask me for credentials and uh credentials are stored um securely. So uh it can do it can use one password um it can use a a secure signin form and so uh this is a super helpful feature just because um credential handling is a sensitive thing. You don't want to be posting your pasting your uh credentials into uh the chat ever. Um, and so by the way, um, Grockbot's

29:12[clears throat] really smart about that. It, uh, makes sure that if you make a mistake, it it flags it. Um, but here, let's see if it gives us a a credential and we'll log in. By the way, once you do this once, you won't have to do it again. Um, I'll just speak a little bit more about that. So, um, uh, our Grockbot can connect to I'll show you that in a second. connect to MCPs and using your credentials and act on your behalf much the way that other AI tools can if you've used cursor before. You've seen this before. Um, but

29:46it can actually navigate uh novel tools. So, if you need it to um figure out an interface or figure out a new um let's see, I'm actually I'm just going to check here because I did this uh yesterday and I Let me see if I've got it open on another bot so we don't have

30:07to sit here waiting. Okay, here we go. Okay, so it was actually able to sign in. So I did it yesterday on another bot and so was actually able to use those credentials and it gave us a screenshot of uh it signed into the application. So this is on its machine. If I open up the computer, I'll see uh I can take control at any time and I can see that it it did indeed uh log in. So we're good to go now. So, I'll go back to recording a task

30:38and I'm going to hit start recording. And so, uh, I'll show this, uh, this here, and I'll I'll open this up. And this is the cheapest, uh, flight. It's $76. And so, I'll go over here and I'll, um, go down to competitor fairs. and I'll add a new one here. Um, so it'll I'll teach the task and so once the uh once I've taught the task and you can also upload your own video, it'll actually watch the full recording. Um, Grockbot can watch videos which is really really cool. Um, and so I I did it here for you yesterday. So I'll show you. So finish the recording. It's like

31:26got it and it it it was able to see exactly what it needed to do. It needs to search the name of the routes, read the chip the cheapest matching fair, open the competitor affairs in 20 which is our our tool and then fill in the route. And so once it's done that what you can do is you can say okay cool you know how to do that now. So run that routine every morning for each of our routes. Uh so this is like the major kind of um evolution uh and and kind of like the the big gain that we get from AI is previously if we wanted to build a

32:02system to do something like this we had to write like a deterministic script and so it was not adaptable. You actually it just executed like a machine. The cool thing about AI is it's able to um adapt a routine to novel circumstances. So for example, if we start shipping new routes, we don't have to update a script. It knows to go like check what routes we have. If there are cases where the fairs, you know, are the exact same, it knows how to handle that. It can navigate nuance. And so we can trust it to do this job. And I just said here like look out, you know, 1 day, 7 days,

32:4114 days, 30 days, and 90 days out. And so it now runs uh weekdays at 7:30 a.m. and it'll do this once a day. Um which is just super helpful. So, okay. Um, so I'm going to go back to uh up here to my new bot, competitive Carl, and I'll say, "Make sure the three new bots report directly to you, never to me, and only surface what I need to know once a week on Monday morning, unless there's something extremely urgent that needs my

33:18attention." And so here I'm going to get a Monday morning report. I do this with my own Grockbot account. Every morning I have uh my report ready and also I have drafts ready in my email. So um rather than having to go in and check my inbox myself, for example, um Grockbot already got there. And not only is it it's not going to summarize, what it's going to do is it's actually going to go and do all of the work behind the scenes to, for example, gather information for a customer, uh, look at updates, verify things, and actually draft responses for me. So, I no longer go to my inbox. I

34:01actually go to my drafts folder, which is super weird, but that's actually what I check every morning is my drafts folder because I know that the things that are important for me to reply to are there and the work is largely done. And so I go through the draft and I make sure that you know sometimes I'll I'll tweak the the phrasing a little bit but like 90% of the work has been done

34:20automatically for me every morning. Okay. So um it now knows uh and it has informed the bots not to contact me directly. And so I'll be getting this uh moving forward every morning uh on Monday. All right. So um that was kind of like a a marketing uh use case. I want to show

34:42you all a support use case. And so um we have here the I I had my Grockbot working on this this morning. So we're going to do it live and see if it works. But um if not, I've got a backup. Um but uh often times we have uh cases where there are a lot of um support issues coming in from different sources. So you might have a support form uh on your website for for your product or for your company. You might have uh an email that people reach out to. Maybe people uh blast you publicly on X when there's an issue to get attention. Um maybe uh you

35:26have a support line. And so, um, one of We're going to try this here. I'm going to call and see if this works. Fly low air support. How can I help? Hi, my name is Santi Garza and um, I'm trying to book a flight and checkout keeps failing at the payment step. Nice to meet you, Santi. Could you give me your booking reference and the email

35:54on the reservation? Sure. It's f ly 49213 and my email is santi.garzaample.com. Your form is all filled in. Please check it and press submit and fly air support will email you an update.

36:21Okay, so you just saw a lot there. Uh for one this was handled entirely by an AI system. Um our I I mentioned earlier that our our uh speech-to-pech voice model uh Grock voice is currently uh number one on on most benchmarks and as you as you saw there um it's incredibly solid. Uh we actually use it at um at Starlink we h we have 20,000 uh calls a day that are handled um largely by our our voice API model. Um, so, uh, I believe 70% of our calls actually get handled 100% by this model and which is super super helpful, right? You don't have to stay in a queue, uh, which

37:11drives me crazy. When I call a company, it gets handled right away. And, uh, the 30% that don't get handled entirely by, uh, our our Grock voice API, it's actually by design. These are cases where we want to speak to our customers. For example, sales calls or things like that. people are inquiring to buy um we escalate those up to our sales team for example um or in really specific cases we um will will kind of bubble it up to our support team um but I highly encourage you all if you have a customer support line or even if you don't to hook it up to uh our Grock voice API

37:46give it a go uh this was actually hooked up by Grock Grockbot last night while I slept so Grockbot can actually hook up your Gro voice API uh which is super helpful. You don't have to set this up yourself and it's part of the outer loop that's really helpful about Grockbot is because it's so well integrated and it's able to navigate Nuance, it's able to set up systems like this for you and manage them. So, um the other cool thing you saw there was here on the site, it actually filled in the information uh live, which is really helpful if for example, if there's a misspelling or you

38:21know something like that, if I want to add additional notes as a caller, I can do that here. And then I can uh submit the form. So ticket was received. Um the the cool thing about this too is uh it'll actually get handled by a by a Grockbot. So let me see if I can find

38:41it. Um one second. Where's support? There it is. So you see support agent is moving around. That means that support agent is working. It says handling it. So the call came in. it was submitted and um for demo purposes I had like a submit button but it could just be a phone call and then a hangup. Um and our Grockbot is actually on it. Super super

39:06cool. Um it knows how to handle these. The way I set this up for anybody who's interested, well Grockbot set it up, but the way Grockbot set it up is it used a web hook. So as soon as your form gets submitted, it actually kicks it off to a bot. And you can you can very easily set this up with with Grockbot. No technical experience needed. You'll notice that the ticket uh was created here. So Santi called about uh uh about um this booking uh number and the checkout kept failing at the payment step and it actually believe it or not uh kicked up a cursor

39:42cloud agent. Cursor cloud agent as a reminder is essentially an engineer on its own computer that has a copy of the codebase. that's able to do engineering work. And so, uh, the cloud agent is going to attempt to reproduce the checkout issue and fix the cause, add a test so that it doesn't happen

40:00again, and then open a pull request. Pull request for anyone non-technical in the room, it's just a coding change that's going to get merged into the product. And it's even going to give me before and after screenshots, and it can do screen recordings, too, uh, of the changes. uh then that'll be waiting for an engineer to review. And this all happened with zero uh need for human involvement. This is support tickets happen all the time. Bugs happen all the time. And so Grockbot and and cursor together are and Grock voice are able to navigate this system and uh that way by the time the engineer sits at the desk

40:41they it's it's essentially done. Um, so, uh, you'll see here, I'm just gonna open this up. Cool. So, you see here, this is cursor.com/agents. And you'll see that this agent is now working on its own. I didn't send this prompt, this prompt was sent by Grockbot. And not only did Grockbot send the prompt, Grockbot is actually able to uh, like it's t it's tapped into this entire agent trace. So, it's actually uh managing and uh keeping eyes on this cloud agent. Um you'll notice the cloud agent actually spun up a um a a sub agent that is uh reproducing the bug. And this is again all happening

41:26automatically. Um like I said, I had Grockbot set up this entire system while I slept. I hadn't even tested it before the demo. Very irresponsible. But that's the that's the kind of trust that I now have in Grockbot's ability to pull this off. So, um okay. Uh I already ran this uh last uh or it ran this last night and it um drafted this PR. So, let's check it out. So, fix the checkout payment

41:56failing. Uh like I said, this is a PR. It's on GitHub. Um it's for coding changes. So if you have issues [snorts] with your uh website, this is probably what's going to manage it. So for anyone technical in a room, uh I know you're very familiar. Um and so here you have a screenshot of the payment could not be processed. Um so it has a screenshot of the issue is able to reproduce. It actually um drafted repro steps. So it

42:22knows exactly what reproduced the issue. It uh I've set it up to write failing tests. So it'll fa fail the test and then it'll um make sure that the test passes when the when when the um change is done. And here you'll see the screenshot of after its changes are done that it now works. I'm still amazed by this and I work with this but like it's incredible that AI is able to handle this. Um I'll just say a couple more things about cursor real quick. I know we have some engineers on the call. Um, cursor also has bugbot which will review your changes to see if

42:59any bug new bugs got introduced. And I have autofix on so it's able to fix uh any issues with the PR on its own. And you can actually set up other te tests here like you can have it uh a few guards at the gate. You can have a security reviewer so that your changes um get checked for all the security

43:17vulnerabilities. You can have a an approval agent and you can even have um this follow the changes all the way out to your production website or your application to check how it affected um production. It's hooked up to your logs and it'll actually uh follow the change all the way out. Okay. Um I'll go a little faster. I know we're short on time. Um I'm going to How are you doing with the

43:47Q&A? uh doing great. I think we have a couple of like fundamental questions, but um u wouldn't take too long. Okay, what I'll do real quick is I'm going to instead of going into Grockbot and showing you this live, I'll just tell you some great uh sales um use cases. So, Grockbot doesn't just help with efficiency and cost savings. It

44:13also helps with with sales, too. Actually, our sales team uh loves Grockbot. Um they finally have a fun AI tool and um and so uh enterprise sales uh sellers spend a lot of their time um a lot of their weeks working on around sales. So researching accounts, updating Salesforce, writing follow-ups, prepping for calls, chasing information across Slack, gong, etc. So, um, but they they only spend they only get to spend a fraction of their time actually selling the product because there's all of this work that needs to be done. And so, Grockbot can live in these tools. So, I'm just want going to show you one

44:53thing here. So, if you go down here to connect apps, you'll see, for example, if I go down here to the filters, um, this is just a super helpful resource actually to get you your ideas going. I'm going to go into the sales section. And you have Granola for meeting notes. So all of your transcripts end up here if you have granola set up. And you can hook up uh Grockbot to Granola. I have this uh for my personal account. We've got Clay here. We've got Gong. So if you record your calls, etc., etc. And you can even um connect your own uh custom uh connectors as well. So once Grockbot is

45:31hooked up to all of these systems, it's able to pull from them. you can um ask it to do one-off tasks, but more importantly, you can have it set up like recurring tasks. So, for example, I'm going to give you a few a few ideas. Um one is to find leads. So, look for buying signals, uh funding news, job posts, leadership changes, and then you build a researched uh target list for each rep every morning. That's a great use case for Grockbot. Another one is writing the first touch. So you can have it draft personalized outreach from each account's research and then and then the rep will review and send it. Another one

46:09is to prep reps before every call. I do this too. Um just a quick thing uh I had a call that I I didn't even know was booked uh while while I was asleep. I woke up and Grockbot had already I had never been with this customer. Had researched all of the previous calls with the customer. had actually done research around the things the customer was asking about or was going to ask about, had prepped a setup uh sorry um a prep uh one pager for me and had even uh created a demo and run a demo on its own machine and verified that it worked

46:43custom to the customer. So when I woke up and I was like what is this meeting this morning? It had already done all this work and it really it saved me. So it can do work on its own and keep an eye out for you as well. It can help keep deals moving. So, flag stalled opportunities, miss next steps, quiet

46:59champions, etc., and draft that nudge. Um, I'll give you another fun one. Uh, you can have it prepare a two-minute podcast for you every morning. So, if you're a sales rep and you want to have a little just while you brush your teeth or something, um, you know, uh, the state of all the customers, etc., etc., that's like a really fun use case. Um, I could go on and on. Um, so yeah, that's those are some sales use cases. Uh, do

47:23we want to open open it up for Q&A? Yes, that sounds great. I think a question that's come up repeatedly is how do you think about like creating your Grockpot team? Uh, like why are you for example creating a separate bot for each workflow versus just having one Grockpot go and do multiple things? Yeah, that's a great great question. So, there are a few reasons for that. Um, for one, having when a when a bot has a focus, it um it performs better. Uh, and and the reason why is because it's context. So there's I could get a little bit into the details here, but there are

48:03essentially two types of memory. There's long-term memory and there's short-term memory. And when a bot has long-term memory around its job, it's able to focus entirely on, you know, kind of that vertical slice of of the uh of the bot or the other cool thing is me as a user personally, I I there are other AI products that that I've used where because they're doing so much, it starts to get really difficult for me to go and check in on them. I would have to scroll like 70 conversations up to try to find the last time that we did XYZ. And so this for me is like really helpful as a

48:46human organizationally to say I know exactly where to go for like my, you know, my [clears throat] competitor analysis. Like I can go in there, for example, if I go to my where is it? Social pulse. I can pull up and I know I'm going to be able to find here's the customer sentiment on how customers are feeling about competitor airlines for example. Do you see how quickly I was able to find that? Imagine if I had to scroll like way up on a conversation. So there are a lot of reasons. Oh. Oh, I'll I'll mention one more. Um they're able to work in parallel. So sometimes there

49:21are 20 bots doing a job at the same time. Each they're doing each of their jobs for example on a Monday morning and they're able to parallelize the work that way. Um yeah, no that's great and I think that's been one of my favorite aspects of Grockot as well. Uh and then a second question, how do we think about shared bots? Like when should we think about individual bots and shared bots for team collaboration and how does the memory

49:46layer for shared bots work? Yeah, great great idea. So or sorry great question. Um so here if I go up here to share um I can actually share my I can bring my team in. So shared bots work alongside your team. You can invite teams to use it. And so I can invite specific people. Um or I can invite everyone on my team. Um on the admin side, you can actually push bots out to specific groups. Um, and this is really, really helpful because for one, you know, if there are a hundred people on your team that all need a particular bot, I think it's a

50:24little silly to try to, you know, to to force them to make their own version of the same thing over and over. Um, but also, um, we didn't have time to get into it. We do have like a an SDK for this. So if you want to actually iterate with your bots um and you know have like um version control for your bots and and kind of like evaluate your bots teamwide, you can do that there too. Um but yeah, I I think it's just helpful um for there to have a team managed bot that uh you can count on to do work for each individual person on your team. Uh

50:58and I'll mention um one more thing. So we have if you go here to bot templates uh we have a marketplace. Uh these bots come out of the box. Um I'll give you an example. So this is Hagglebot. Um what Hagglebot did for us is it found is the number $100,000 I think is the public number um of yeah of savings uh for like unused seats and licenses. It haggled um vendors to bring our our cost

51:29down and it did all of this on its own. Uh it was so useful we actually made it public for everybody to use. So you might get a huge cost savings here. All you have to do is import it and I'll just show you what happens. So you have context. So here it it has kind of its its base um system prompt. It has memories of like what it's supposed to do and these come out of the box. It also has routines. So it has two routines. So for example, monthly inventory refresh um and it has integration. So it's hooked up to ramp to Slack to notion to Google Drive. And

52:04so this is just one example, but we actually have quite a few uh bots in the marketplace that you might find really really helpful. That's great. Uh and then you already answered another question uh which was like how do we see uh like Dropbox's memory or context or its system prompt for example and you already like showed us that flow and I think we maybe have time for one last question is how do you think about like creating end-to-end workflows with Crockpot? Like do we have to be very specific and like have one bot created uh like how do you just like

52:42start with that process? Yeah, great question. So our philosophy here is is essentially crawl, walk, run. So first crawl, then walk, then run. And what that means is you want to start slow and you want to start hands-on. So think about hiring a new hire. If I create a bot, I there are a few things I need right out the gate, right? First is I need to be specific about what I want the bot to do. So bots cannot read your mind. And so they're actually incredibly helpful out of the box. They are great at inferring what you want, but if there's something very specific you

53:17want, you have to communicate that. And so kind of day one with a bot or you know, you know, first 10 minutes are should really be about trying to lay out uh the focused work exactly how you want it. Then what we like to do is in the first few days, you really want to hold its hand and iterate. you know, you kind kind of constantly want to um reinforce the things it's doing right and and tweak where it's doing things in in a way that you don't want it to. And this helps like um build up trust, right? So, it has long-term memory. It's going to

53:51remember your preferences. And so, this is like a really really helpful part of the evolution process for a bot. Eventually, you know, you want to start like spot checking. So, maybe you don't check in on it every hour or whatever. you check in on it maybe once a week and see the kind of work that it's doing and you just kind of lay your foot off the gas more and more until it's running fully autonomously. And that's how we like to build out our workflows. Um but uh yeah, you also have the the uh the agent SDK which you can look up uh we have docs on that that helps with hill

54:24climbing so it'll improve itself. It's got eval out of the out of the gate. So um that's very helpful too. Yep. Uh, I know we're at time. Um, but thank you so much everyone for joining this call. Your questions were super insightful and Sandy, thank you for that killer demo. We actually had a lot of people asking if the visualizer is actually in the product. Uh, which is pretty incredible. But again, thank you everyone for your time. We really

54:49appreciate it. I'll I'll end by saying that the visualizer is something you can have Grockbot build for yourself. Uh my brother who works here, Lucas, actually built that visualizer and we all use it internally, but you can build your own. Just ask Rockbot to uh to show you its workflows. All right, thank you all for joining and uh have a great day or night wherever

55:12you are.