Let's start this talk with the origin story of your journey as a founder. You have to go way back to middle school when you were reading the essays from PG, right?

Early on, I had been interested in starting a company for a long time. I had been interested in a bunch of other things too. Originally, I got into programming because I was interested in starting something commercial. The first time I ever saw code was over some winter break. My brother and I wanted to create a hit mobile game. We didn't really know how to do that. We looked on Google: how do you create a game? We heard that you need to download this application called Xcode. We did that and we were hit with these weird, colorful, esoteric symbols which were Objective-C, which is still around but maybe a little bit less popular than it was then, for good reasons. We stared at this impenetrable wall of Objective-C and my brother promptly ejected. He didn't move on with programming. He now is on a very different career path. He's trying to paint or something like that. But I kept going and bought a book on Objective-C and then eventually started working on mobile games. That was the genesis of me getting into programming.

Along the way, yes, I was a big fan of PG's essays and Sam's essays too, and a bunch of the folks in YC. That was definitely a big inspiration even from the very early stages of high school.

The wildest thing about Cursor is that right now you're just 24 and built this monster of a company in a really short amount of time. To a lot of people it could seem that it's a bit out of nowhere, but this was really in the making for more than a decade. You've been working and shipping a lot of different projects, right? And you were working in AI even when you were in high school, right? Tell us a bit about the projects and how you got started with that.

I was lucky enough to find programming early on. I was also lucky enough to be interested in AI early on and have some great collaborators to work on AI projects with. Soon after the foray into mobile games — I wasn't very good at mobile games — so one of the things that I built, and one of the things that got most popular, which was the technically easiest thing to build (maybe a lesson in startups: the code isn't everything), was this mobile app where you could spoof high scores in things like Piano Tiles and Flappy Bird and then send them to your friends. That was the thing that went viral. It wasn't the painstakingly handcrafting the game engine yourself thing.

Soon after that, I got interested with a friend in the idea of building a robotic dog where we thought it would be really great to have a robot that you could teach to do things without programming it. Instead, you could give it positive and negative feedback like you give a dog. So you could give it a treat if it does something good. You could say "bad" if it does something bad, and then maybe you could teach it to play fetch and things like that. That idea really animated us. But we had no idea how to build it. And so again, we started where one would start, which is Google, and went down a lot of rabbit holes. That took us into learning about genetic algorithms, and maybe that was going to be helpful for building this robot dog that we wanted to build. Then we eventually learned about neural network stuff because some people were playing with taking genetic algorithms and using them to evolve neural networks at the time with work like NEAT. Eventually, it took us to reinforcement learning, which even back in 2015 people had been working on for a long time.

In the end, my friend and I did eventually build a couple of robots. We didn't do any sort of substantial work that really lasted, but we did work that was interesting at the time in taking reinforcement learning algorithms and making them more data efficient, making them better at learning from very, very few data points — on the order of tens of data points — and also from noisy data: data that a human's giving. It wasn't exactly a dog, but we built a couple of robots where one of them was this many-axis robot arm that could swing a paddle and play ping pong. If you put the right sensor on it and then you gave it the right sort of positive and negative feedback, you could teach it to swing when it sees a ball. Then we had this Kiwi drive robot that we would teach to follow a line.

It was a great education in ML, partially because of our dumb naive approach where we didn't really know that there were things like Torch and TensorFlow and lots of other building blocks we could use. Maybe we weren't good enough at Googling.

So you like implemented your own neural network from scratch?

Yeah.

When you were like 16, 17?

The constraints of the problem were we were dealing with robots, and so we were dealing with microcontrollers. Microcontrollers have very little memory and they couldn't really fit any of the normal standard ML libraries. So as part of our bike shedding trying to build a robot dog, we implemented our own tiny neural network library. I have memories of us not really understanding any of the internals of how these things worked or not really understanding calculus, but fumbling our way through re-implementing some important ideas from neural networks. I think it taught us a lot. There were a lot of gaps in the fundamentals that it took many years to fill in later.

Then fast forward to the founding of Anysphere. It's an interesting name because Cursor is not what it is. When you guys started, you had just graduated MIT, right? That was back in 2022. What was the first idea that all four of you started working on back in 2022?

The genesis of Cursor was in 2021. My co-founders and I had been interested in AI for a long time. Each of us had our own little robot dog moment. One of my co-founders worked on trying to build a competitor at Google using LLMs in 2021 and training his own contrastive models. One of my co-founders worked on computer vision in academia, and some of us also worked on recommendation systems at companies like Google. We were really interested in AI. In 2021, we were trying to figure out what to do with that interest. Do we go and work on AI in academia, or do we join a big existing AI effort, or do we start our own thing? There were two moments that really got us excited. One was seeing the first AI product start to come out — GitHub Copilot was really the canonical example for us. The other was seeing work about how it looked like AI was going to predictably get better in the future as you scale up these models. At the very beginning of 2022, my co-founders and I went on a month-long hackathon basically, and we started hacking on ideas related to picking an area of knowledge work and building what it looks like as AI gets more and more mature.

You guys have collected a lot of data for that first idea, right?

Yeah. The first real idea that we worked on for a long time was in mechanical engineering. We were trying to build a co-pilot for mechanical engineers and trying to train models to predict what you would do in a CAD system like SolidWorks or Fusion 360, which is where mechanical engineers model out parts in 3D on a computer. We picked it because we thought it would be boring and sleepy and uncompetitive, and we were doing an armchair MBA thing even though it was a horrible choice from the get-go because none of us were really mechanical engineers and also the science wasn't really ready for that area.

But you guys kept working at it for a number of months, right? And you crawled and got all these CAD files and actually got something working with autocompletion, right? That was like the first version of it working.

Yes. A bunch of the work was in data scraping. Honestly, it was trying to get all the CAD models on the internet. There are also all these different file formats and trying to convert them all into something that's canonical because CAD is this weird software market where there are all these different systems that are pretty popular. It's very fragmented. There are also cloud CAD systems that don't have easily exportable files and they don't want you to scrape their stuff, so there was a bunch of work there. Also, the training infrastructure for doing any kind of modeling work back then was pretty rudimentary, so there was a lot of work on the infra side there and just a lot of experimenting with models and a lot of experimenting with how you even jerry-rig an extension into these CAD systems because we were building an extension. These applications aren't really extensible at all.

There were also other projects that we were working on at the time. Two of my co-founders were working on an end-to-end encrypted messaging system because one of them has a background in security research. The idea there was apps like Signal and WhatsApp encrypt the body of the messages, but they don't hide who's talking to who at what time, which is actually really crucial information if you don't want to trust the messaging app provider. So if a journalist is talking to some informant in the government, just knowing that they're communicating at all is actually a really big piece of information.

So, that was in the middle of 2022. You guys were working for about a good 6 months on this idea. How many users were you getting at that point?

All of these projects were elevated and had basically no users.

At what point did you realize that the idea was not working? It's like, oh no, we're all working on this. We're trying to do the startup. It's not working. And what was that moment like?

It was a bit different for each of the projects. For the messaging system that two of my co-founders worked on, it was really technically impressive, but it had these bad trade-offs where it wasn't very scalable. They tried to give it to people and it didn't really work. Then they tried to sell it B2B and it didn't really work. For the CAD ideas, it was after many months of trying to get the models to really be useful for end users and then also reckoning around: are we really interested in these areas, or is there something else that we're inherently much more excited about?

So there was a moment that you decided "okay, these ideas are not working, we have to pivot again." You turned through three, four, five ideas before landing into code completion?

Yeah. We had been inspired by tools like Copilot really early on, and we had avoided working on AI and coding because we thought it was too competitive then — still is competitive now.

I think so, or more, potentially more.

And you guys are like, "Oh, we could still do a better job than GitHub Copilot because people thought the game was done."

Well, we didn't think we could at the start. Then it was the desperation of having worked on ideas for a while and not really being excited about them after a while and them not really working out. That shapes what you care about and what you're aiming for. We realized we were really inherently excited about the future of coding. We also got to see how some of the other people in the space were working on their products. We got to see how the tech was developing. We took a step back and realized that if we were being really consistent with our beliefs, there was going to be an opportunity for all of coding to change in the next 5 years and for all of software development to flow through models. It felt like no one working on the space at the time was really taking that seriously. It felt like they had great products and they were making them a bit better, but they weren't really aiming for a world where all of coding as we know it today gets automated and building software ends up looking very different. Then with that in mind, we set out to work on that.

That was a bold move because you said "okay, we're going to stop working on all these other ideas that we didn't have as much of a background in" and you were excited about programming even though you had this big Goliath in the room with GitHub Copilot. You decided to go and just solve this problem.

It didn't really feel bold or like a move at the time because it's a bunch of people sitting around in their living room on laptops. It's not like pivoting some giant company, but we did. Initially, we waded into it where we were thinking, well, maybe we do this very niche tool for basically security reviews — trying to detect future CVEs in your code — or maybe we build something that's just for this one niche area of software. We thought about building for quants and prototypes and things just for quantitative researchers. But in doing that, we were just brimming with ideas for what Cursor could be if it were just about trying to be the best way to code with AI in general. We had a ton of conviction about that and we had a ton of excitement about that, so at some point we just decided to go for it.

And that was end of 2022, when you decided to make that move, right? How quickly did you ship the first product and what did the first product look like? You shipped it a couple weeks later — what did that look like?

It did take us a little bit of time to ship something publicly. It took us roughly three months from first line of code to open it up and G it. Originally, what we did is we built our own editor, quote-unquote, from scratch. It was still using a bunch of open-source building blocks. There are a lot of great primitives like CodeMirror and the language servers, and there's a lot of open-source tech that can help you build an editor, but it was cobbled together from scratch. There was our own version of remote SSH or our own Copilot integration at the time because we didn't have anything like autocomplete. You have to build your own peen system, you have to build all your own language server integrations. There's just a lot that ends up going into something as developed as the code editor market, making something that can actually be competitive there and serve as someone's daily driver. It was four weeks until we built something that we could use as our daily driver. It was maybe four weeks later where we gave it to the first beta testers, and then there was another four weeks and then we G it. It was still very crude at the time. It didn't feel like a big thing to just open it up to the public.

What did you learn in that first version? Because you built a code editor from scratch. You guys haven't done the whole forking yet.

The other thing is you guys had also implemented your own models too. Like back then you got a lot of inspiration from Codex, right?

Yes. When we were setting out to work on our first idea that we really spent a bunch of time on, which was trying to help mechanical engineers be more productive using AI, one of the things when we raised our first round of funding was because we actually needed money from the get-go to do a little bit of model training because you couldn't bootstrap it with the models that existed off the shelves. They weren't good enough at that task.

One of the papers that we would tout around is actually the original Codex paper because by our calculations, Codex, which was the first autocomplete model behind GitHub Copilot, didn't really cost that much money to train even though even back then at the beginning and middle of 2022 people were talking about how expensive AI models were to train. My math might be wrong, but I think it was about 100k in training costs. During this foray into mechanical engineering, we had done our own training, and then when we set off on Cursor, we were a little bit burned by that. So we wanted to be as pragmatic as possible, not to reinvent the wheel. We started by doing none of that. Then over the course of 2023 in dialing in the product, that ended up being a really important product lever, especially as we got to scale and we got a bunch of people using the product. That also gives you the ability to use product data to make the product better. That actually has been a really important muscle to build in the company.

What happened then in 2023 was when you were still not sure about whether Cursor was going to be a thing, right? You were still debating with your co-founders whether you should still pivot. It's like, oh, is this idea still going to work? And you were still trying to grow it, right? Because it took a long time to get to revenue, right?

Yeah. Over 2023, it was growing. The numbers were kind of small, and we were working on something where there wasn't always a clear next step. There are probably some markets where you're really well served by going immediately talking to a bunch of people, listing down their problems really rigorously, or really systematically and exhaustively thinking through each problem: what would be the direct solution, and then prioritizing them and then going from there.

We were and are in a space that's a bit different than that. We're this end-user application that doesn't have much of a complexity budget. We are trying to build the best way to code with AI. A lot of that is figuring out, given the tools that you have today, what can you actually do? There's a lot of things that you could write down that would be useful if you could build them, but then figuring out how to build them and all the details — it's not entirely clear how to move forward on that. There were a lot of times over the course of 2023.

Also, to add to this, of our early user base, if you just followed the gradient of exactly what they wanted, you would get pulled in slightly different directions than we ended up in. We had a really loud segment of users that didn't know how to code at all, and we talked about: should we focus on those folks? We had a really loud segment of users that wanted us to do things that were very tech-stack specific — just building for one technology and making it much less of a horizontal tool — and we resisted doing that too.

There was a lot of early prototyping and wandering the desert in 2023, and then figuring out things around: where does it make sense to not just build the software but also build our own models to improve the API models or to replace them in places, for instance with our tab, our next edit prediction, and then how exactly to do that.

You went from zero to 1 million around 2023, right? And it took a lot to get there, right?

Yeah, it was a bit more than that, but sort of roughly that.

And then 2024 was a crazy year. You guys went from one to 100 million in one year. Tell us about this loss of compounding power because you kept that growing 10% week over week. How did that happen?

The numbers felt small early on, then the compounding kept going. There were a couple of things that really drove our growth. We're in this market where if you make the product better, you see it in the numbers immediately — things start to grow more. We felt it around when we first started to make Cursor codebase aware, when we first started to be able to predict your next action, when we made that more accurate, then when we made that faster, then when we made that more ambitious — it could predict sequences of changes — and then when we let the AI model start to take more action within your codebase and then do that really fast, speeding that up. All along the way, we just focused on making the product better, and the compounding continued. I don't think this is true of all markets, but we're in a market where end-user preferences matter a lot. If you make the best thing, people hear about it and talk about it. That kept going for a long time.

One of the funny things that happened around that time: we did see a big shift in the YC companies as they were going through the batch because we would ask what kind of tech stack you use to build your applications, and it was night and day from one batch to the other. I remember in 2023, it was maybe single-digit percentage of the batch would use Cursor. Then 2024, it was like 80%. It just spread like wildfire. The best builders were using you.

Got onto their Twitter feed. Yeah.

The Twitter feed. Was that where a lot of adoption came from? How did all the growth come from?

The very early stages when we were first launching the editor, we tried to evangelize it on social networks. Actually, one of my co-founders, when the dopamine hit was keeping him going in 2022 when we were working on some of these ill-fated ideas, started posting on the internet and explicitly set out to try to gain a lot of followers not by doing normal social media things but by talking about AI. It was surprising the degree to which someone could actually just read all the papers, think deeply about what was going on at the time, talk about that publicly, and then get recognized by influential people in the space. There was this particular open-source model, Flan-T5 at the time, that multiple AI efforts that ended up using that model. They found out about the benefits of that model directly from my co-founder just because he was posting on Twitter and doing that consistently.

He became like a sort of niche, very niche, sort of niche niche niche SF micro-celebrity. He would evangelize the product early on. We had this very movie magic demo. When we first launched and when we first did a wait list to just get our initial batch of users, that was helpful getting us kick-started. But then after that, we stepped away from that and we lived like monks in 2023 and just focused on the product. It really just spread from word of mouth.

I remember there were a couple of times during that year where there were members of the team that would say things like, "Guys, the product's already good enough. Let's put it aside. Let's just focus on growth engineering." And then there would be a two-month sprint on doing some version of that, and it just never worked. It kind of washed away compared to the other stuff that we worked on that year.

And by that time in 2024, how big was Cursor? How big was the company at that point?

It was pretty small in 2023. My co-founders are fantastic engineers, and there were four of us, so we could go pretty far without hiring anyone. We also had our own set of missteps in figuring out the first set of people to hire and how exactly to do that. We were both very patient early on and also focused on hiring a lot less than we probably should have early on. I think we ended 2023 at only single digits people. We were less than 10 still. Yeah.

Amazing. Shifting gears a little bit: what are your thoughts in terms of how the future's going to look with coding?

We were this maybe middle-road bet from the start. When we set out to work on the company and we were hiring our first people, we would get these weird looks. At the end of 2022, it wasn't really like this because ChatGPT happened and then the whole world woke up to things beginning of 2023. But especially during 2022, when we were working on the CAD stuff and then the early code stuff, people thought working on AI was kind of weird to do. People were not entirely convinced that it was a good use of time and that there were going to be lots of great applications to fall out of AI. Even the people who were interested in AI in our space — there was a bunch of people that were just focused on optimizing the form factor that existed already and just making those products a little bit better.

At the same time, in our social circles and professional circles, there's a bunch of people that were thinking, oh, why would you work on anything other than AGI? All of the work that you're doing right now, in one or two years, circa 2022, is going to go away. We've always had this view that there's going to be lots and lots of incredibly valuable things to build over the next couple decades. AI is going to be this transformative technology. Maybe more so than any technological revolution in recent centuries, but it's going to take a couple of decades. It's going to be this industry-wide effort where there are all of these independent capabilities that each need to fall into place to really get to a place where you can entirely get to the end state of transforming building software on computers or the other areas of knowledge work that might be transformed by AI.

Concretely, in the near term, we think that for professional engineers, which is the end user we serve, the market that we serve, code is still really important. There will be this long, messy middle where you will be working with the AI more and more. It will become like a colleague more and more. It may also become like a very advanced compiler that can start to hide some of the code for you. You're going to have to read the logic and review it and edit it.

So, what do you think are the skills that are still going to matter? What should everyone still be studying or stop studying?

I think that programming, like math, is just a good general education. I don't think that goes away. There's also lots of practical skills that come from studying computer science right now. Often when people are entering dynamic industries, the specific stuff that they study in school isn't super crucial. It's more the learning that they get along the way. I don't think that's changed with AI.

What advice do you have for the audience? If you have a young Michael Truell, maybe not just three years ago, they want to be like you three years ago before you started Cursor. What should they be doing right now?

I think just working on things that you're interested in and doing it with people both that you enjoy being around but that you respect a ton. And taking that really seriously. For a lot of people that are in school, there are so many things that pull you toward more checking boxes and less focusing on building something up over time and really focusing on something that you're interested in.