# How Cursor's Browser Experience Shaped Grok Bot: Nick Miller Interview

Channel: Callstack
Video: https://www.youtube.com/watch?v=2Hi7GVb1yqo
Duration: 11 min
Language: English
Words: 1871
Transcript page: https://viewrankai.com/tools/youtube-transcript/2Hi7GVb1yqo

---

[0:00] Hey everyone, uh this is an interview recorded at Agent Conf 2026 in Warso conference uh created by Colstack about AI software development. I'm here I'm Mukashinski. I'm joined with Nick Miller from Space X AI. We just watched your talk, me and Luke. We are pretty excited. So we invited you for a interview. Hello. Thanks so much for

[0:21] having me guys. Honored to be here. Can you just say maybe how the last few months of your life looked like? Who were you before you got to the stage as SpaceX spokesperson? Yeah. Uh I mean it's been a bit of a blur. Um I'll tell you since I started at uh Cursor uh the whole thing has been a blur. The speed at which that company has grown and just this space in general has moved. Um so the last 3 months have been really wild. A new chapter in the company. Um you know we started with this um initial relationship with SpaceX AI uh where they were interested in

[1:00] acquiring us and we had this period where um you know they were going through their IPO. Um we were kind of feeling each other out. Uh they hadn't officially exercised uh their option to acquire us. And so the way the relationship started was that um they gave us access to cosis and uh allowed us to start uh training uh on that system. Um so we have been working on model models for really most of the lifetime of cursor. Uh the first model uh that we offered was our autocomplete model um tab uh which is really kind of the first model in the space that I think really got

[1:38] autocomplete dialed in. And um since then we have been working on models but the dream was always to build the best model for writing code and just like any uh company that is working on models you know one constraint was access to compute. Um so what was amazing with glosses is that it feels like almost unlimited compute and really opened up uh what we could do in terms of running different

[2:03] experiments working on different models. So we finished the training of composer 2.5 on Colossus and then we started collaborating with the SpaceX AI team uh on Gro 45 and then Gro 46. So that was kind of the early days of the relationship. Uh and then I think it's been about 3 weeks uh since they

[2:24] officially exercised. I know I could be wrong. As I said it has felt like a blur. Um but it has really been incredible. um they have been very warm and welcoming and I think are excited to uh bring us on board and um you know when I joined this company I thought we were just building uh a tool for writing code we're launching rockets and potentially putting data centers in space and so it's been incredible learning about everything that SpaceX Space XAI does and you know what their mission is or I

[3:00] what our mission is. Yeah. So maybe like on this subject, how how does it work when two companies combine and then like I'm sure you have your own workflows at Cursor and XKI, they had their own like inner workings of how they train their model, how they use their model, how they iterate on their kind of a products, right? So then when you merge, you have your own composers in the background, then you have Gros, then you have the the other models for SpaceX. How does that work in terms of the synergy? How does this work in terms of okay so we are leaving that one behind because it

[3:38] doesn't get so much value for us anymore but and we are focusing on like these two things two coming from XKI maybe and one coming from Corso yeah you know I think we're still feeling that out but one advantage has been that I think we've had pretty welldefined swim lanes so you know SpaceX AI has a launch business where they're putting rockets in space and putting tonnage in space and then they have Starlink

[4:01] famously yes and then on the uh AI side. Um you know they have a suite of APIs. Um they've got a great voice API. Uh they've got a great uh image and video API. They have which didn't. Yes. Exactly. And then on our side um you know we have all these great coding tools and they do have Grock build uh which is a CLI tool that's really really quite good. um we have a CLI but I would say that we put a lot more emphasis in the desktop app cloud agents that sort of thing. Um so I think we've complemented each other really well and um there so there we haven't had to I

[4:41] think address as much of that. I think we're looking at how the brands are going to work together and kind of what a unified platform looks like. Um, you know, we've got some customers that are signed up to cursor plans and then also XAI plans and kind of figuring out how those are going to come together. But, um, you know, as as big as their company is and as fast as that we've moved, um, it has gone pretty seamlessly considering um, yeah, so it's I think that has really helped though that we've kind of had pretty well- definfined swim

[5:10] lanes. You also mentioned something uh, before which is Colossus? Yeah. So, Colossus is one of the biggest data centers in the world for training models. Um, so you know, large language models are really the first time in technology where the more money you throw at something, the better it is, right? If we're working on a software product, um, and we we have unlimited budget, um, you we could hire a thousand engineers or 10, chances are we're probably going to do a lot better with 10, right? Yeah, but with training a model, it's like the diminishing returns,

[5:48] right? But with training a model, the more compute and the more training data that you can throw at it, the smarter it's going to get. Is it linear though or is it also like flattened out at the end there? It has not flattened out yet. Uh so that is the question is when will it or will

[6:06] it flatten out? Um but it has not yet. So um the labs that are creating these models are continuing to find more and more training data. And I think something that's interesting about SpaceX AAI is that uh and now with the acquisition of cursor is that I think the Grock models are unique in that cursor has more training data on software development than any other company in the world. Um SpaceX AI has all of this data around SpaceX and the LA has all this engineering uh work X as well. Exactly. And then with X, it is all this real time data whether it's around sentiment or news. So it gives

[6:45] our models uh definitely some some some competitive advantage I think in in some of the ways that it can serve. But what's exciting about Colossus for us is that we've always had these model aspirations. Um but just like every other lab we are trying to find enough compute and now with access to Colossus it I don't want to say it's unlimited compute but it has allows us to do so much more than we could before you know before the acquisition um we were working on our composer line of models that were really purpose-built just for writing code and one advantage of that was that we didn't need as much training

[7:21] data to train those so we could do it faster we could do it for less money once we got access to Colossus. We started working on the Grock models. These are general purpose large language models. They're trained on everything. Um, and we were able to go after those ambitions now because we had so much more compute and then that opened the

[7:39] door for Grockbot. I know that you have been using Grogbot since its launch. Yes, Grockbot is uh phenomenal. And one thing like I wanted to get your input on is like how is it so good compared to like uh claudin uh chrome or clothes computer use and I've even tried to for example set up computer use with like something with like versel eve um nothing is comp nothing is comparable to what crockbot can do in terms of computer use I mean like if you try to do it with eve or cursor it's getting blocked by different websites. It's running into issues. It's not being able to complete the task. But with Grockbot,

[8:21] you give it pretty much anything which uses computer use and it does it. So, I'm wondering like how did you guys get the computer so good? I'd love to hear that. You know, um I think we we really benefited from the fact that we have several years of experience building Cursor and building great harnesses, uh building computer use in Cursor. you know, we launched a built-in browser to the cursor desktop app uh in the last year. Um, and so, you know, we we built Grockbot from day zero to an internal beta in about a month. I mean, it's pretty astounding. And and this company runs faster than any other

[9:02] company I've ever been at. But I think the reason we were able to do that or one of the reasons is because we're building on this corpus of knowledge that we have developed building the cursor platform. Uh and I think that carries over to the browser use stuff also. I think also because uh Grockbot is you know your family of bots have their own virtual machine and it's really siloed and isolated. Um you know we're working in a controlled environment there. Um, but we knew that that browser use was a really, really critical part to making this great, right? We wanted the bots to be able to

[9:36] be as powerful whether there is an API or an MCP server available or if they're just going to interact with a browser just as a human would. And I think it will continue to get better, too. Yeah. And uh for the examples of what can you do with Grogbot, go to our YouTube channel, watch Nick's talk from this conference when he explains in a really simple language, in a really simple like life examples how you can use this to organize your life better or how you can use it in your professional life. And with that, I want to thank you for being here in Warso at Agent Conf

[10:12] 2026 speaking with us and uh yeah, excited to see what's comes next out of XAI. Thank you guys so much for having me. I really appreciate it. It's been a pleasure. [music]

---

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