Decoding Cursor's Rise: How a VS Code Fork Became the AI Coding Powerhouse in Three Leaps

In the AI coding space, Cursor has been the brightest star of the past two years. Starting as a fork of VS Code, it has grown into a platform adopted by over 50% of Fortune 500 companies — completing a three-stage transformation from IDE to agent platform to model platform in just two years. This article, based on a retrospective conversation with one of Cursor's early investors, attempts to reconstruct the strategic choices and cultural DNA that drove the company's ascent.

Back to the Beginning: A World Where AI Coding Was Still in Its Infancy

To understand Cursor's victory, you need to rewind to late 2023 and early 2024. The leading models at the time were GPT-4, Claude 3, and Llama 3 — yes, Llama 3 was considered quite impressive back then. In the coding assistance space, GitHub Copilot was the consensus winner by a wide margin.

"Microsoft was doing what it does best," the investor recalled. "It had Copilot, Office, and VS Code — put those together and it seemed destined to win AI." Concepts like Agents, Loops, and Reasoning that are commonplace today simply didn't exist yet. AI coding had shown promise, but it was an entirely different world from what we know now.

In that environment, startups fell into a few camps: one camp believed "the only way to win at coding is to train a specialized coding foundation model"; another focused on agents; and hundreds more were building plugins. Cursor chose a path in between — not training its own model, not building a plugin, but forking the entire IDE.

The Contrarian Bet: Why Forking VS Code Was the Right Call

In hindsight, investing in Cursor seems like an obvious decision — but at the time, it was a deeply contrarian bet. The competitive landscape was almost "absurd": a company called Microsoft, with scale, OpenAI's model weights, and control over VS Code and a hundred million developers. By conventional wisdom, no startup could survive against that single competitor.

But the Cursor team had a coherent internal logic. On the question of "why not build a coding-specific foundation model," co-founder Michael's answer was: You don't talk to these models in code — you talk to them in natural human language. Therefore, models must understand the full breadth of human experience and language, which means building a coding model is really building a frontier language model — an extraordinarily difficult undertaking.

Their other core insight was: Future code will look like pseudocode. Michael often referenced early computer science papers that "described algorithms in pseudocode rather than code." His view was that in the future, you'd only need to compress a programmer's intent into the smallest possible specification, and the model would implement it. This prediction was later proven entirely correct.

The investor recalled that during the pitch to the partner team, Michael spent 90% of his time saying "no." VCs would throw out questions like "would you consider expanding into related areas?" and Michael would listen politely, then say, "Interesting, but no — we're not doing that." This rare level of focus and clarity became a key source of the team's conviction to invest.

Product-Led Faith and Breaking the Rules of Growth

What made the Cursor team uniquely distinctive was their identity as a thoroughly "product company" — which is rarer than it sounds in AI. They held a firm belief: if the product is good enough, it will be adopted. This belief drove nearly every decision they made.

The investor compared this mindset to Musk-style conviction: if you truly believe in your product, you'd never build a plugin, because that means becoming part of someone else's product; if you truly believe in your product, you also wouldn't rush into enterprise sales, because the product itself is the best go-to-market strategy.

This extreme product conviction led Cursor to shatter many conventional assumptions in growth investing. When a growth investor suggested "typically you bring in a sales lead at $25M to $50M ARR, when self-serve growth starts to plateau," Michael looked them in the eye and said: "Self-serve growth hasn't plateaued at all." That statement forced the investor to reexamine the assumptions baked into every growth model — you simply cannot apply a standard 25% growth trajectory to a company whose growth curve keeps accelerating.

From Series A to Series D, the timeline was compressed to an extreme: the lead investor closed the Series A in May–June 2024, and by October, the Series B was done. The growth curve was described as "vertical takeoff."

Staying Calm in the Face of Giants

Cursor's entire growth story has played out under the constant shadow of "imminent existential threats." From Copilot to Windsurf, from Cognition to the first version of Claude Code, competitors have cycled through in rapid succession.

In May 2025, Claude Code arrived with considerable fanfare. A few months later, the investor asked Michael for his take. Michael's response was memorable: "We're targeting the largest market in the world. We've had rotating competitors from the very beginning — Copilot was the first, and Claude Code is just the latest in that series. In a massive market, you'll always face strong competition. That doesn't scare us."

The investor described this attitude as "a blend of humility and confidence" — reminiscent of Bezos — the kind of composure that comes from being smart enough and right enough about the big things to face giants without flinching.

That said, the team did have a few moments where the pace of model evolution genuinely stunned them. "I remember when a frontier model dropped and we were all on vacation. Everyone was in awe — these models are evolving so much faster than any of us imagined."

Self-Disruption: Three Pivots in Two Years

Cursor's most underrated capability is its Reed Hastings-style willingness to cannibalize itself — decisively and without hesitation.

They started with an IDE, then shifted to an agent platform, then shifted again to a model platform. Tab autocomplete — the feature that inspired Karpathy's famous "tab tab tab" post — is no longer the centerpiece. Very few founding teams have the discipline to proactively retire their own core product features.

This also explains Cursor's later "back-door strategy": acquire as many users as fast as possible, then use the accumulated assets, data, and know-how to build their own models. As the investor noted: "You can't reverse-engineer this — not unless you started as a frontier lab." This strategy drew heavy criticism at the time, but ultimately proved to be the right path.

On the Twitter discourse that summer about Cursor's gross margins, the investor offered a sharp take: "Technological transformation always precedes the establishment of business models. Think about the early internet — it took a long time to figure out monetization, let alone margins. Here, you have unlimited demand, unlimited growth, and room to expand margins."

Cultural DNA: From a Love of Coding to the Fastest Sales Team Ever Built

Cursor's early culture grew from the team's pure love of writing code and building systems. When the investor asked Michael what he did outside of work, Michael could barely comprehend the question — and when he finally admitted to having hobbies, those hobbies turned out to be, unsurprisingly, writing code.

But there's a paradox here: if you look closely at what the team actually spent time on later, they poured enormous energy into recruiting, go-to-market strategy, understanding business models, and business development. High-energy founding teams are always doing whatever the business needs most.

On the sales front, Cursor executed what might be "the fastest sales team buildout in history." They took the methodology they had refined for engineering recruiting and applied it wholesale to sales hiring: first, study the top 10 companies best at a certain type of sales; then drill into the top 10 teams within those companies; then identify the key individual salespeople who drove those transformations; and then conduct layered reference checks. The founders spent up to 40% of their time on recruiting — exceptionally rare for technically-oriented product founders.

This philosophy of "how you do one thing is how you do everything" permeated the entire organization, even showing up in office design: Nordic Hygge aesthetics, slippers at the door, carefully curated secondhand sofas — all designed to make the whole team feel as immersed as they would at home.

Acquisitions and Integration: The Most Concentrated Acqui-Hire Playbook

Cursor also excelled at acqui-hires. They never worried about operational complexity — "no matter how messy, we'll figure it out" — and they never doubted their ability to set the cultural tone, because they were exceptionally good at it. Early acquisitions were pure team grabs; it wasn't until the Graphite acquisition that they began thinking about M&A from a strategic direction standpoint.

The investor highlighted Cursor's "playbook" for integrating founder-caliber talent: Tito from Koala, Adam Ward on the talent side — both were able to slot perfectly into Cursor's culture and elevate it to new heights. While not unique to Cursor in the generative AI era, Cursor is undoubtedly one of the best examples of this approach — they didn't just plug these people into individual contributor roles; they genuinely used this method to build the company, with an extremely high degree of operational complexity, and they pulled it off.

A Startup's Magnum Opus

Starting by competing with Microsoft, eventually going head-to-head with giants like Anthropic; completely evolving their product path and go-to-market approach within two years; executing highly complex M&A integrations on top of all that. The investor called it "a startup's magnum opus."

Cursor's story shares many parallels with SpaceX: both survived being written off countless times, both aimed at the largest possible market in what seemed like impossible competitive conditions, and both won. That cultural resilience in the face of skeptics may well be the defining trait of the best AI startups of this generation.