Inside Cursor: The Anatomy of a Generational Startup

**Martín Casado** (0:00)
We don't need to compete with Anthropic and OpenAI on models right now. The interface between the human and the model is the key thing.

**Matt Bornstein** (0:07)
If you looked at the competitive landscape, it was almost silly.

**Sarah Wang** (0:12)
And he said something to the extent, look, we are going after the biggest market in the world. You're always going to have formidable competitors. That does not scare us.

**Martín Casado** (0:21)
As founders, we all want to be ambitious, and we want to push to the maximum point of the Pareto frontier, but you have to know what the curve is.

**Matt Bornstein** (0:27)
For us, the decision to invest in Cursor was actually pretty obvious. You remember, Andre Carpati was using it. It was clearly a phenomenon. It was clearly like a known brand.

**Sarah Wang** (0:35)
They went and got all the users in record time. Now they have the asset, the data, the know-how to build their own models. You really couldn't do the flip of that unless you started as a frontier lab.

**Matt Bornstein** (0:46)
They were definitely a couple of old moments. Like remember one?

**SPEAKER_4** (0:50)
Two years ago, betting on an independent AI coding company looked almost irrational. Microsoft owed GitHub, VS Code, Co-pilot and had access to some of the world's best AI models. And yet, Cursor broke through. In this episode, Martin Casado, Sarah Wang and Matt Bornstein look back at the decisions that shaped Cursor from its earliest days, building a standalone product instead of a plugin, resisting the push into enterprise sales too early, and staying relentlessly focused even as the competitive landscape changed around them. They also unpack how the company repeatedly reinvented itself as AI advanced, built a culture around speed and taste, and eventually applied the same intensity it brought to product to hiring, sales, and M&A.
It's a conversation about Cursor, but also a broader look at what it takes to build a defining company in a market where the technology, competition, and conventional wisdom change almost overnight.

**Sarah Wang** (1:53)
So to kick off, I actually wanted to take us back to early 2024 Obviously, you guys led the deal in, I think it was May of 2024 that year. But can you walk us through what was going on in the Q1, Q2 of that year and maybe how you guys started working with Cursor well before we ever partnered officially with them?

**Martín Casado** (2:17)
You have to transport yourself back to late 2023, early 2024 I went back and looked it up. The leading models at the time were GPT-40, Claude 3, and Llama 3

**SPEAKER_4** (2:26)
That feels like forever ago.

**Matt Bornstein** (2:28)
Llama 3?

**SPEAKER_4** (2:28)
Yeah.

**Matt Bornstein** (2:29)
That was actually pretty good.

**Martín Casado** (2:30)
That was great. Yeah, it was a great model. Change in strategy, obviously. Now, the leading coding harness was Copilot by a mile. This was the consensus thing. Microsoft was doing their Microsoft thing. They had Copilot, they had Office, they had VS Code, all of these things combined. They were going to somehow win AI.
It was clear that AI coding was working, but it was a different universe compared to today.
You had the early signs that you could do meaningful work with AI, but agents didn't really exist, loops didn't really exist, reasoning didn't really exist at the time. It's a crazy time.

**Matt Bornstein** (3:05)
There was a continuum of companies going after coding.
On one hand, the only way to win coding is to build a base model, and it has to be a coding specific one. On the other hand, you had very appreciably, by the way, companies just focus on agents, and Cursor was somewhere in between.

**Martín Casado** (3:25)
Not only models, there were a million companies doing plugins too. Both of these were very reasonable things to do done by very smart people.
That's what's so interesting about these new spaces. It's like smart entrepreneurs show up and do really interesting things. In this case, model training was a very reasonable thing to do because the models weren't that good at coding yet. There was a theory that if we focus on this and do a coding-specific model, that's the breakthrough. People doing plugins on the other side were saying, hey, let's not toss out the whole IDE. That would be a crazy thing to do. But the interface between human and model is really important, so plugin is a great place to go. It just turned out that the thing that Cursor was doing was the right thing to do.
Like us, they were very bitter lesson-pilled, but at the time, they would have said, oh, there's no need for us to go train our own model. Of course, that changed later for different reasons, which we're going to get into later in this podcast. But it's like we don't need to compete with Anthropic and OpenAI on models right now. The interface between the human and the model is the key thing. This is something that I remember Michael and Eman saying a lot, and just saying very clearly from the earliest days of Cursor, saying code in the future will look like pseudocode. I remember Michael making this point a lot. Like when you look at old computer science papers, here's the algorithm not in code, but in pseudocode, which I'm sure many people are used to. We don't exactly write in pseudocode now, but probably you could. Probably you could just take a pseudocode block from an old paper, and the models now would implement that. This is really what the Cursor guys sought from the very beginning, and they've proven to be exactly right about this. It's this natural language-ish specification, pairing the programmer's intent down to the minimum possible spec. That's what really matters now. Anyway, we were in this world, as Martin said, we were talking to a ton of very smart, very interesting people working on coding in various ways. The Cursor team stood out because not only did they align with the things that we thought about the world, more importantly, smarter people than us and engineers working. At the time, the leading AI companies like OpenAI, Mid Journey, Replicate, a bunch of others, all use this and subscribe to this model. And the team frankly just seemed kind of special, right? They had an unusual degree of focus. I remember we got Michael to come pitch our GP group and 90% of the meeting was him saying no to things, which is just like for founders out there working on products that you love, like VCs will always ask you kind of dumb questions about other things you might do or things you're related to or whatever. And Michael literally just sat there and like very politely listened to all the questions and said, hmm, interesting, no, we're not gonna do it.