No Model of Their Own, No Early Enterprise Sales: Winning on Interface and Speed
Cursor's success did not come from training its own model but from treating the product itself as the moat: no plugin, no early enterprise sales, no in-house model training — it came from behind on iteration speed, a hiring philosophy, and founder density.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The future of code is pseudocode, and the interface is the moat
From the beginning Cursor said explicitly that it did not need to compete with Anthropic or OpenAI at the model layer — the human-computer interface was what mattered. Michael and Amman made the argument repeatedly and very early: code in the future will look like pseudocode, compressing the programmer's intent into the smallest possible specification, with natural language as the new way of programming. That judgment is what put them on an independent-product path instead of building a plugin, and it determined everything that came after it strategically.
There is no coding-specific model; it is just a frontier model
Asked why they were not training a coding-specific model, Cursor's argument ran like this: humans speak to models in natural language, so a model has to understand the full breadth of human experience and language — which means building a coding model is really building a frontier language model, and that is simply too hard. Building a plugin, meanwhile, means letting your own product live parasitically inside someone else's. They believed that a product good enough would get adopted on its own, so no plugin and no early enterprise selling — a highly consistent chain of product-company decisions.
Self-serve has no ceiling, so the old growth-stage model breaks
In the traditional growth-investing model, self-serve softens once a company reaches $25 million to $50 million of ARR, and that is the moment to bring in a sales leader. Michael instead looked investors in the eye and said self-serve had not topped out. That forced them to discard historical assumptions like ‘growth decays to 25% five years out’. In the end they were not held hostage by conventional growth metrics and dared to make the bet — which is exactly where they were later proven most right.
Microsoft held every ingredient and still lost on iteration speed
At the time Microsoft had VS Code, the OpenAI weights, an enterprise sales machine and 100 million developers — very nearly every ingredient there was. Cursor's opening was speed: John Schulman said Anthropic's monorepo died the moment it loaded, and the next day Cursor had fixed it and could show him. Iteration that fast proved that writing this as a plugin on top of VS Code fundamentally does not work — the smartest people had already tried it and failed. History has also shown again and again that in a large market the independent player wins in the end.
Copilot was only the first; the giants will keep rotating in
After Claude Code's high-profile launch in May 2025, investors asked Michael what he made of it. His answer: we are chasing the biggest market in the world, so there will always be competitors — Copilot was just the first, and Claude Code is one of many rotating cast members. A big market guarantees formidable rivals, but they do not frighten us. That blend of humility and bravado is close to Bezos-style confidence: if you are smart and you are right, a giant opponent does not get to set your tempo.
Technology shifts come before business models, so the margin doubts do not hold
In the summer of 2025 X filled up with posts about Cursor's gross margins, yet the business kept doing well. The historical pattern is that technology transitions arrive before business models, and investors picking at the economics of a new technology have always been proven wrong. Cursor's strategy was to capture all of the users first, at record speed, accumulate the data and the know-how, and only then build its own models — a path that, taken in the other direction, only a non-frontier lab could walk. They took a great deal of abuse for it, and it is being proven right.
Hire salespeople the way you hire engineers: back-channel each one
Cursor recruits AEs exactly the way it recruits engineers: first work out which 10 companies in the industry are best at this particular kind of selling, then get deep inside those teams, find the number-one and number-two AE, and back-channel them repeatedly until it is confirmed. What the early stage needs is people who can cope with uncertainty, not people who can only run a mature playbook. They put the same meticulousness into sales hiring, which is how they expanded quickly to more than 50% of the Fortune 500 without diluting the culture.
Founders do not belong in IC seats; use them to build the company
Today everyone thinks of Anthropic as the place with the highest founder density in AI, but Cursor was practising it earlier: not stuffing founders into IC roles, but genuinely using them to build the company. Tito came from Koala; Adam Ward runs talent. These founder-CEOs amplified leverage fast once they were inside. The model is operationally extremely complicated, but Cursor proved it can work. In the war for talent, the companies that understand talent-via-acquisition will pull far ahead.
In their own words · checked verbatim
We don't need to compete with anthropic and open AI on models right now. The interface between the human and the model is the key thing.
Code in the future will look like pseudocode.
It turns out that you don't speak to these models in code. You speak to these as humans in natural languages. Therefore, the model has to understand the breadth of human experience and language
Michael looks me dead in the eye and is like, self serve is not.Petering out.
Look, we are going after the biggest market in the world.There's going to be competitors, in fact there's been a rotating cast of characters.
first comes the technical.Transformation then comes the business model.
Figures
| Cursor's Series A | May 2024 | 1:11 |
| Leading models in early 2024 | GPT-4, Claude 3, Llama 3 | 1:11 |
| Developers Microsoft had | 100 million developers | 14:11 |
| Cursor's Fortune 500 penetration | over 50% | 25:00 |
| Share of founders' time spent hiring | 40% | 22:01 |
Glossary
- self-serve
- Users sign up and pay on their own, without going through a sales team.
- back channel
- Verifying and sourcing candidates through private networks rather than a public hiring process.
- talent via acquisition
- Acquiring companies in order to bring in outstanding teams.
- monorepo
- Keeping all code in a single repository, a common practice at large companies.
How to listen
AI application founders, early-stage investors, growth leads, and every developer still stuck on ‘plugin or product?’.
The closing disclaimer read-out can be skipped (roughly after 38:30).