# Building a Software Factory with Cursor, xAI, and Grok Bot | Nick Miller at Agent Conf 2026

Channel: Callstack
Video: https://www.youtube.com/watch?v=iHmBg0N7EOA
Duration: 27 min
Language: English
Words: 5084
Transcript page: https://viewrankai.com/tools/youtube-transcript/iHmBg0N7EOA

---

[0:00] [music]

[0:14] [music]

[0:23] Great to be here with you all. Um, my name is Nick Miller. I am a uh lead field engineer at SpaceX AI. Uh still getting used to that. I I uh worked at Cursor and came over from the acquisition. Um and you know when I first started putting together uh the talk for the session um several weeks ago, uh it was a very different talk. Um but some things very exciting things have happened here at SpaceX AAI in the last month or so uh with the release of Grockbot. And so what I really wanted to focus in on today was Scrockpot and um how it is really changing the way that

[1:02] we work whether you are a software engineer or an executive or on go to market. Um the problem we're solving with with grabbot is that most meaningful work I think is bottlenecked by wrote reactionary work. um in our days uh I think that we are regularly getting emails, we're getting Slack messages. There are all these things that we have to react to and act on and that often carves into the time that we can really do the meaningful work uh the special work that really moves the needle forward. Um I was an early employee uh at Kurser. I was employee number 26. Um before that I founded a

[1:46] SAS company uh that I ran for about 12 years. And uh any of you here in the room that have worked at an early stage startup, I think you know the feeling that I had in the early days of cursor in the early days of my SAS company where you are just overwhelmed with work there. The ratio of the amount of work that you have to the number of people that you have working is way off balance. Um and uh what Grockbot does is it acts as a knowledge worker. It allows you to offload a lot of that reactive work uh that repetitive work and frees

[2:24] you up to really do the meaningful work that is most enjoyable that is leveraging what you are great at and again is is what really moves the ball forward. Um Grockbot was released uh on August 11 so about a month ago. Um we actually at cursor uh we built this product in about 4 weeks. Um it started out as just an experiment and I think after about 10 days we realized that we had something really special. Uh I personally have been using it for about 2 months. It was in an internal beta uh at at cursor at SpaceX. Um and I will tell you if they took it away from me

[3:01] today I don't know what I would do with myself. Um I I've started referring to it as crackbot. Actually when I talk to an enterprise uh I'm just like get the executives Grockbot. We don't even need to do a session. You just need to start using it. Uh and you will not be able to put it down. Um so over the last three years uh you know no one has really cracked the code I think on making a singular interface that is really useful across a wide variety of domains. Um, we first started with chats, right, where you were able to essentially ask the chatbot to do something, it would give

[3:38] you some information back and then you would act on it. Uh, and then we got into the co-pilot area era. Uh, cursor was a part of that as far as software goes. Um, you had other companies that were releasing legal agents, finance agents, teaching agents, all of these different agents. But I think the idea of this general agent framework was still uh left unsolved. Um think about Grockbot is really a new agent primitive. Um these bots are not limited to just retrieving information or just one knowledge area. uh bot bots can really function as a team of knowledge workers with the ability to take an initial directive and turn it into

[4:22] routines that accomplish actual units of work for your organization. Uh the bots can complete really complicated long- form tasks that these early primitives were not capable of. I think that is one real difference here is that you can really give a bot something that is many many turns of work and it's able to finish it end to end and I'll give you a quick example. I'm going to talk a lot about using Rockbot in a professional uh context but I use it personally as much as I use it professionally. I'll give you a quick example just to kind of wet your appetite. Um I travel around the

[4:58] world for cursor for SpaceX AAI. I'm currently actually living out in Singapore for about 3 months helping us spin up our APJ operations there. And uh my wife and my kids are out there with me. Uh they joined me a few weeks ago and uh it's about a week before uh I uh they were arriving in Singapore to join me and we still didn't have a place to live and my wife was freaking out understandably so. Um I was super busy with work and I I needed to find us a short-term rental. [snorts] So I had Grockbot do it. I told Grockbot, uh, this is what we need. We We need a

[5:34] three-bedroom furnished place. We need it for 3 months. Um, uh, I have two daughters. One is 15. Uh, her boyfriend was coming to visit for a week. There was no way in hell they were going to be in the same bedroom, so I need an arrangement where he could be siloed away and locked up at night. Um, I gave Grapot all this information. I said, "I want you to go out and research and find uh short-term rentals for us, and I want you to create a Google sheet of all of those rentals." So, I gave it access to my Google Workspace. It went out, it researched, it came up with about 35

[6:07] different properties that met my requirements, put together this sheet, then I said, "Email this to my wife." Now, this is one of the first lessons I learned is that it emailed it to my wife uh in my name. So, it signed kind regards Nick at the end. My wife was like, "What the hell is this?" She knows that every email I send her, I never put any sort of signature in there. That was one thing we had to correct with Grockbot. But she then went ahead and she marked five different properties that she liked. She highlighted them yellow. She told me when she was done and then went back to Grabbot and said,

[6:38] "Okay, now I want you to go ahead and I want you to schedule appointments for me to view these appointments." Now, Grabbot is also connected to my calendar. So, it was able to see what days I was available when I had windows of a time to go see these places and then it actually contacted the properties and set up the uh visits. I went out, I visited the properties, we narrowed it down to two and then Grockbot actually negotiated the leases with both of these places and got us all the way to the finish line. So, very different from the co-pilot era and the chat chatbot era in terms of how wide

[7:13] that endto-end work can really go. Um ease of use I think is is really important. Um when we look at a lot of the existing knowledge work uh products out there I think that there have been some challenges when we've tried to kind of find the perfect form factor. Um they're really complicated to set up. Uh they run locally on your machine. Um there are all sorts of limitations

[7:39] around connectors. They're flaky. They're hard to configure and you're constantly having to babysit them. And it's just not easy to share these things. It can take a lot of time to get them set up and then it's not easy to kind of democratize uh that access and provide it to other people. Grockbot was really designed to be super simple. In fact, you know, one of the mantras we have at cursor is that uh is delete the product. Uh we are as you know if any of you use the cursor platform, we are shipping new features all the time. This

[8:12] is a space that's moving extremely fast. Um, but we are deleting features as much as we are adding features. So, deleting the product is really an important kind of pillar of how we work. And we definitely took that approach with Grockbot. If you were part of the internal beta that we had with Grockbot before we did an external release, there was actually a lot more interfaces, a lot more that you could do with it. And

[8:35] we actually trimmed a lot of that away. So, it is really just dead simple. It looks like an iMessage uh sort of interface. So it's designed to be very easy to use. When you open up the application, it just it just works. Um so when you define a new bot, there are kind of three parts that you want to focus in on. Context, connectors, and outcomes. So as far as context goes, when I'm designing a bot, uh I always start by giving it a name and uh some sort of initial understanding of what it does. You can see in this example, we

[9:11] have a bot that's called sales outbound. And the bot will even look at the name and automatically make some assumptions about what its purpose is and then ask you sort of how you want to use it. Then connectors. I think this is really kind of the magic of what makes these bots so powerful. You are able to connect them to all of these outside sources. So at cursor for example um there are maybe about a half dozen kind of information

[9:37] sources that are really key to my job. There is Slack which is really kind of the nerve center of our company. Uh there is notion. There is Google Gmail and calendar and docs. Um there's Salesforce. I use Granola for all my meetings. So I have the bot connected to all of these. And then it has eyes and

[9:58] ears and fingers in all of these tools. Um and then outcomes uh the agent needs to know what it should do. So in the same way that we might measure a human and how many meetings they book or how many PRs they create uh bots need to have a similar sort of goal in what they want to accomplish. So when you deploy a Grockbot for your team uh you know it might look like you're you're working on iMessage all day. I mean this is I I can't tell you how much my day how much my work uh how much my personal life has transformed. I spend so much of my day

[10:30] now in this interface as opposed to in Slack or in notion in these other tools. Um, but I think one of the things that really makes Grockbot special is that it learns from you. So most of the other tools in this space when you ask a bot to do something it is essentially kind of launching this ephemeral session that is there just for that task. These bots persist. So what that means is that they continue to get smarter and smarter and as you guide them, as you correct them, as you give them more information, as you teach them, they get better and

[11:07] better at the work. Um, this is a good graph of of what I was talking about at the beginning uh of the talk here. I think on the left side, this is really a lot of our day. I can tell you this is what my day is. All that red stuff. And honestly, I've had plenty of days where most of those blocks are red. I get so many messages in Slack, email, to-d do. I could spend all day honestly just in Slack and never get to the really strategic high value

[11:37] work, which is why the company hired me. And the idea with Grockbot is really to get the bots to take on as much of that repetitive work. Uh so it can really free you up to do the magical work uh that is most fulfilling and and why you were hired or why you founded your company or why you do your work. Um I think another core differentiator uh with Grockbot is that all of these bots share a computer in the cloud. So, a lot of the other uh knowledge work tools out there um they're running locally on your computer. Uh but all of your bots have a

[12:18] computer in the cloud with its own browser that persists as it works. So for the average user, it means that these agents just just work. You don't need to manage the runtime environment on your local machine. You're not competing for resources. Um if a let's say a non-technical user wants to develop a prototype uh the agent has a place to operate that isn't constrained by the user making the request. Um and this also solves the problem of system integration. Uh giving the bots access to a browser means that they can interface with systems that may not necessarily have an MCP server or an API. So a bot can utilize the computer

[12:58] to use technology to navigate through a browser uh in the same way a human would and this really opens up a world of possibilities. Um Grabbot is the fastest growing product that the cursor team has ever developed. I was looking at the numbers here for this week. 418,000 weekly users. Hopefully that number will

[13:18] tick up a little bit after this session. And more than half of those users are using Grockbot for more than four days a week. Um, I use it every single day, morning, noon, and night. In fact, at 5:00 a.m. every morning, my chief of staff sends me an initial message, which gives me a readout of my day. It's looking at my calendar and telling me what I'm doing that day. It's pulling in information from other sources to help educate me on those things I've got to do. It's looking at my inbox. It's looking at my Slack. It's telling me what are the high priority things I need

[13:53] to kick tick off and then I'm using it throughout the day. So the goal here with Grockbot is really to have it used wherever you do knowledge work. Um Grockbot is a generic agent primitive. Uh bots are not constrained to be only good at one particular domain. Uh they can be a a team of autonomous uh knowledge workers that can transcend all different types of jobs. Um at SpaceX AAI uh users across all parts of the company are leveraging bots to increase their productivity. Um single engineers are having the impact of 10 engineers

[14:30] the 10x number that we always hear. Individual go to market members are accomplishing what previously look took an entire team. And so far feedback has been really positive. Any of you that are on Twitter um I know there's always kind of a flavor of the day or flavor of the week. Uh it is very much a horse race. I think in in terms of uh Twitter with all of these different companies working in AI. Um but we've had a really great moment with Grockbot and I think the reception has been um really positive. I think we really have landed on the proper form factor uh for this

[15:03] type of work. So let's take a look at a few specific use cases just to kind of give you an idea of how you can use this tool professionally. Um executive use cases are really good example. I think that can apply to anybody in this room. I talked a little bit about the morning briefing that I run uh with my chief of staff bot where it is giving me an update each each morning briefing me on my day. And again, it's bringing in information from all of these different

[15:31] connectors uh that I've hooked it up to. Meeting prep is another big one uh that I use this for. So, I am meeting with customers throughout the day. I'll be honest with you, with my aging brain, any of you that have worked with customers, I don't know, you know, you can have these great meetings with a customer and have a very meaningful experience and then by the end of the day, you have met with six other customers and it's all just kind of muddled in your head. So, I'll go into a meeting and not quite remember what we had talked about in the previous meetings. Uh, I the bot is able to

[16:03] surface all that up for me and brief me on what we talked about. Again, it's tying into the granola call transcripts. It's tying into Slack messages that I had with my colleagues about uh this customer. It's looking at Salesforce. So, all of this information. And for me, one of the the first kind of light bulbs that went off with Rockbot is how good it is at searching all of these data sources. I'll tell you personally, I I find the Slack search to be very frustrating. um our notion there is so much information in there it can be hard to find things. Um I found that the

[16:39] first thing that was really powerful that I discovered in Grockbot was that it's really good at search not just from one source but from all these different sources. So the fact that I can walk into a meeting and have a brief there from my Grockbot that is synthesized data from all these different sources and got it set up for me uh can be

[16:57] really really powerful. Um let's talk about engineering. Uh so I know a lot of you are engineers in the room and the original description of this talk was really talking about the software factory. We have a lot of engineers at cursor now that are using Grockbot as their primary surface for building. Um so when you ask uh Grockbot to build something it is going to launch a cloud agent. Uh so that's how it does all of its development work. And um any of you that have played around with Origin, which is our new entry into the source control management space, uh with Origin you can actually start working

[17:34] without even creating a repo, but you of course can use it with GitHub, GitLab, any of the other SEMs. Um but I think something else that's interesting about Grockbot is that you can actually create a team of bots that can cover a lot of different personas uh in an engineering team. So you might have a front-end bot, you might have a QA bot, uh you might have a documentation bot, you might have a research bot. And so you can work with with all of these bots. The engineers that are working on the bot team, they are so botpilled that this is really the primary tool they're using now to uh to

[18:14] build. Um they have moved off of like using the cursor desktop app and um and using using Crockbot a ton. Um so so pretty pretty interesting. It's just such a simple interface and again the fact that these bots can run end to end uh really can be quite magical. Go to market uh any of you that are here uh working in early stage uh startup a founder um even those of us that are engineers have to talk to customers we have to sell and I just I can't tell you how powerful Grockbot can be uh around GTM tasks. Um, and and as I'm talking through these, I want you to know that

[18:51] there is um we have a a guide section on the XAI site um that has guides on how to best use Grockbot for a lot of these different use cases that go really deep, whether it's engineering, product management, GTM. Um, so if any of this is of interest to you, that can be a nice way to kind of get inspired. I'll tell you, for me, um, when I first started using Grockbot, it took me a minute to kind of wrap my head around how to best use it. And we had a session at Curser uh within the first week or two of getting the tool where people

[19:22] started sharing what they were doing. Um we have an account executive at Curser named Christa Less who actually just did a uh a um a workshop that was on Zoom where she talked about how she uses go to market. When I heard her talk about how she was using it like I I cracked the code and it completely changed the way I use Grockbot. So would encourage you to check out some of these guides. I think once you kind of wrap your head around some of the ways to use this um it can be quite magical. One other thing I would say the next big light bulb that

[19:53] went off for me is that the bots can talk to each other. Um which can actually be kind of funny when you see them messaging each other. So I'll give you another example. Um I as I mentioned I do a ton of travel. I'm traveling I've been on the road basically for a year traveling around the world. Um, I spend hours and hours every month booking my

[20:14] travel. Um, I'm a bit of a hotel snob. I'm kind of picky on where I stay. I've got, you know, certain airlines that I want to fly. I want an aisle seat. If I'm in premium economy, I want to be in the front row. Um, I've gotten kind of picky about it. I have a travel bot. I have a flight bot. And I have a hotel bot. My flight bot knows all of my preferences around what airlines I want to travel, what my what class I can fly in, what my seat preferences are. I've got the same thing for hotel, what lines of hotels I like, what's important to me

[20:46] in a hotel. And then I just talk to the travel bot. I say, "Hey, uh, I have a meeting uh, next week in Amsterdam. Uh, you know, I'm going to be in London. I want you to book my travel." That bot will then look at my calendar. It'll see when that meeting is. It'll know the timing of when I need to fly in, when I'm going to fly out, and then it's going to talk to the flight bot and talk to the uh to the hotel bot, and both of those bots are going to work away at getting the flights and the the hotels booked. Uh and then they're connecting

[21:16] to Navon, which is the travel platform that we use, uh either through the MCP server or through the web browser. So, the fact that the bots can talk to each other can be really, really powerful. Anyway, as far as go to market goes, um so many different things you can do here to automate away a lot of the repetitive tasks that you have to do uh in go to market. So, you really can focus uh on how to best be strategic with your

[21:41] accounts and uh support your customers. Um talked a little bit about how bots work together here. Um, you know, a limiting factor I think with a lot of the existing products is that they're very isolated in their context. Um, and in the way that we work in real life, you know, it's very collaborative where we are sharing amongst each other. And as in the example I gave before, um, this is really how the bots can can work together. Um, and bots aren't just limited by your laptop. We have an iOS app. We have an Android app. you know, I can kick something off from my laptop

[22:14] and then I can leave my hotel and while I'm in the Uber, I can check in on my phone and watch it. Again, every bot has its own computer. So, I can close my laptop and it will keep running. Uh, you can also have routines with bots, so they can check things or do things on a recurring basis that is not contingent on uh your laptop or um or your your device. Um, bots, as I mentioned before, are meant to be democratized. So,

[22:39] Grockbot allows for very easy sharing. Uh, once a bot has been created, we have a bot marketplace now where you can install these bot templates. Um, I created a closer bot. Um, so I am, as I said before, I'm in all of these meetings all the time with customers and I'm also often promising them something

[23:00] or multiple things in those meetings. Maybe we're talking about cloud agents and they're asking me about security and I say, "Well, [snorts] we've got a couple of good security docs that outline our security posture uh with cloud agents." Um, but then I've got to remember to send all those things. I've got to write up the email. So, I have a closer bot. At the end of every meeting, Granola sends a web hook to my Grobbot to tell it to analyze those meeting notes and find any deliverables that I've promised. That Grockbot is also looking at my email. It's looking at my Slack and it is creating a queue uh

[23:34] stack ranked of all the things that I need to do. Um and then I can open up that closer bot at any time and let's say that security document I can tap on that or click on that and then it will go ahead and draft the email to the customer with the links. It's tied into all of our documentation. So I don't have to manually write that email. I can go in make a few tweaks hit send and it can go. So using the sharing capabilities, I was able to share this closer bot with the rest of my field

[24:02] engineering team. It's been a big hit. Um so this is a great way that you can kind of seed your bots. Um like there's a great bot called Hagglebot on the marketplace uh right now that you can use to go and negotiate. We used it uh with renegotiating our Zoom uh contract at Cursor and [snorts] uh it saved us like tens of thousands of dollars completely done by by the bot. So sharing can be a great way to kind of seed your bots and get ideas but also share them across your organization. Um the key uh to really building a great foundation with your bots is uh a few

[24:37] different things. Uh giving them access. Uh so using those connectors and hooking them up to all the different things that you do. Uh investing in teaching your bots. Uh again just like when you're using agents let's say in cursor. um setting up sort of guard rails, setting up instructions to supplement the model's knowledge to make them smarter and smarter. Uh so you can just talk to your bot and give it information. Bots can create skills. You can ask a bot to create a skill or update a skill. Uh also in computer use, you can click a button and it will learn based on the way you're working around a browser. So

[25:14] again, these bots get smarter and smarter. And then delegating between bots. the idea that these bots can talk to each other. You can even create chat rooms with multiple bots uh inside a lot of our engineering teams are doing that. So in summary, uh why I think Rockbot is really special and worth taking a look at. It is dead simple to use. Uh I onboarded my 76-year-old mother last weekend. Um and this is someone who, you know, you look at her iPhone and like the text is this big. She's been asking

[25:47] me technical questions my entire life. Um, she is using Grockbot. They're always on. They have their own computer. Uh, they use tools just like you do through a web browser. You don't have to have an MCP or an API. They're really good at finishing complex work endtoend like my Singapore housing example. And you've got sharable templates. Um, so my hope is that out of this talk, if you all haven't tried out Grockbot that you'll be inspired to download it. uh check out some of the guides. Um give yourself a little bit of time to kind of wrap your head around how they work. Um and uh let me know what you think. My

[26:24] email's nick@cursor.com. Um I I I I I you know, I say this in all honesty, this tool, just like cursor totally changed my life a couple years ago, this has completely changed the way I work and has just improved the quality of my life. I mean it has just freed up my time to be a better professional at my work, to be a better dad, to be a better husband. Um I just the quality of my life has really uh improved uh in every aspect. So hope this was eye- openening, inspirational for you. Have a great conference. Thanks so much for having me. And this was my Star Trek

[27:01] moment for sure. Thank you everybody.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt