Context

Three a16z general partners, Martin Casado, Sarah Wang, and Matt Bornstein, look back on backing Cursor (the AI coding tool made by Anysphere) from its earliest round through its growth stage. In early 2024, betting on an independent AI coding company looked close to irrational: Microsoft owned GitHub, VS Code, and Copilot, plus access to leading AI models. Cursor won anyway, and this conversation is a retrospective on why: the product bets its founders made, how growth investors had to throw out their usual playbook, and how the same discipline that shaped the product later shaped hiring, sales, and M&A.

The Big Idea

Cursor's edge was never a smarter model. It was total consistency: one clear thesis about where value would accrue, and the discipline to let every hard decision, product shape, go-to-market timing, hiring bar, culture, follow from that same thesis instead of hedging across several bets at once.

Where most competitors in AI coding were what the partners called "pastiches," trying model training, plugins, and agents simultaneously because they weren't sure what would work, Cursor's founders had, in Casado's words, "tremendous clarity on exactly what type of company they were," and it showed in how consistently their choices held up under pressure.

Key Insights

Betting on the interface, not the model

In late 2023, Cursor's founders decided not to build their own coding-specific foundation model, a path several well-funded rivals took. Their reasoning: developers don't actually write to these models in code, they write in natural language, so a coding-specific model still requires solving the much harder problem of building a full frontier language model. Cursor founder Michael Truell's framing, repeated from the company's earliest days, was that "code in the future will look like pseudocode," meaning the real product surface was the interface pairing a programmer's natural-language intent down to the minimum spec, not the model underneath it. That thesis also ruled out building a VS Code plugin: Casado noted that "if you really believe in the product, you would never do a plug-in, because then you're part of somebody else's product."

Growth investors had to abandon their own playbook

Sarah Wang described pitching Cursor on adding a sales leader once self-serve growth typically starts slowing, the conventional trigger being around $25 to $50 million in annual recurring revenue. Truell rejected the framing outright: self-serve wasn't petering out. Wang had to throw out the standard growth-model assumption that revenue growth eventually asymptotes, since Cursor's own trajectory, by her account, went from roughly $4 million to $50 million in annual recurring revenue in about four months at one stage. "You can't write growth memos that way," she said, meaning the usual comparable-company framework simply didn't apply.

Competitors are a constant, not a crisis

The partners listed a rotating cast of "formidable" rivals Cursor faced over two years: Microsoft Copilot first, then Windsurf (which built momentum through YC's developer community), Cognition's agent-based approach, and eventually Anthropic's own Claude Code. When Claude Code launched in 2025, Casado asked Truell directly what he thought. Truell's answer: "We are going after the biggest market in the world... we think in big markets you're always going to have formidable competitors. That does not scare us." The partners described this as humility mixed with bravado, and contrasted it with how most founders visibly react to a new competitor's launch.

The team cannibalized its own product twice

Cursor moved from being an IDE, to an agent platform, to a model platform, each shift replacing what had defined the product before. The "Tab" autocomplete feature that Andrej Karpathy had publicly praised (a moment the partners credit as a real signal in their decision to invest) later became a minor part of the product as the company moved on. Bornstein compared the pattern to Reed Hastings deliberately cannibalizing Netflix's own DVD business, calling it something "very, very tough for founding teams to actually have the discipline" to do, since it means retiring what made the company famous before a competitor forces the issue.

Enterprise sales entry was timed to where the margin was

Early on, Cursor's only enterprise motion was a generic inbox, "if you're an enterprise, send your inquiries to this," and Truell explicitly turned down suggestions to hire a sales leader. Casado's explanation for the eventual pivot: across the large AI labs, the self-serve tier is often subsidized or given away, while the real margin sits in enterprise and third-party deals. Once Cursor's leadership concluded that's where both the spend and the margin were concentrated, they moved fast: Wang credited the team with building what she called one of the fastest-growing enterprise sales teams she'd seen, reaching over 50% of the Fortune 500, using the same intensive, back-channel-driven method they used to hire engineers.

Mental Models & Frameworks

Tech transformation precedes business model

A pattern the partners use to counter margin and business-model skepticism during a major platform shift. Casado's version: "First comes the technical transformation, then comes the business model," pointing out the internet didn't have a working monetization model for years after it was clearly transformative. Applied to Cursor, criticism the company took over its gross margins in mid-2025 came right before Cursor kept growing anyway, since public skepticism about margins tends to arrive before, not after, the business model catches up to the technology. The caveat: this explains why early margin critique is often premature, not that margin stops mattering once the technology is proven.

The backchannel method for hiring outside your core function

A way to transplant a company's product rigor into a function its founders don't have direct expertise in, illustrated by how Cursor hired its first account executives. Instead of running a generic search, the method works in layers:

• Identify the 10 companies known for excelling at the specific type of selling the role requires.
• Within those, identify the 10 best individual teams.
• Within those teams, identify the one or two specific reps who actually drove the results, verified through multiple layers of backchannel references (their boss, their boss's boss).
• For an early-stage hire, weight the search toward someone who can operate under real uncertainty, not someone who only knows how to run an existing playbook.

Sarah Wang called the underlying idea "how you do anything is how you do everything": the same intensity Cursor's founders brought to product also showed up in how carefully they built the recruiting engine, spending roughly 40% of their time on it even as technical founders.

Trade-offs & Nuance

Talent acquisitions versus strategic acquisitions

Cursor's approach to M&A changed as the company matured, and the partners were clear the two modes require different judgment. In its early acquisitions, the company optimized purely for bringing in strong teams and didn't worry much about operational complexity or whether the acquired team would absorb Cursor's culture, since the team's existing strength made that manageable. The Graphite acquisition, described as much larger, marked a shift toward acquisitions made for strategic direction rather than talent alone, requiring more deliberate integration planning. The trade-off: moving fast on talent-only deals works when team quality alone justifies the risk, but a larger, strategy-driven acquisition needs the operational planning that early deals could skip.

Common Mistakes

Mistake: reacting to competitor headlines instead of your own thesis

The partners contrasted Cursor's founders, who they said were "100% unfazed" by a string of well-funded rivals, with how most founders behave when a competitor makes news, visibly anxious and prone to reactive pivots. Truell's response to Claude Code's launch (grounding the reaction in market size and long-term thesis rather than the immediate headline) is the pattern the partners point to as the better instinct: treat a new competitor as expected evidence you're in a real market, not as a signal to abandon your own reasoning.

Practical Application

Write down the actual reasoning behind a product bet

Before committing to a product direction (build versus integrate, standalone versus plugin, when to enter a new channel), write out the specific mechanistic reason for the choice the way Cursor's founders could always give what Casado called "a very reasoned articulation." Then check whether your next few roadmap decisions still trace back to that same reasoning, or whether you're layering on unrelated bets because the original thesis feels shaky.

Locate where the real margin sits before staffing a new motion

Before building out a new sales or growth motion, map where competitors in your space actually earn margin, not just where growth is happening. Cursor waited to invest in enterprise sales until the data showed that's where spend and margin were concentrated across the large AI labs, rather than defaulting to a sales hire at a standard revenue milestone.

Use layered backchannels to hire outside your expertise

When hiring your first specialist in a function your founding team doesn't have deep experience in (an early AE, a first recruiter), apply the same rigor used for technical hiring: name the peer companies known for excelling at that specific work, identify their best teams, then their best individual performers, and verify through multiple levels of backchannel reference before extending an offer.

Schedule a retirement date for your own signature feature

Pick the one feature your product is currently best known for, and set a recurring review (quarterly, at a major roadmap checkpoint) to ask whether it should be de-emphasized or replaced as the market shifts, rather than waiting for a competitor to force the question. Cursor moved from IDE to agent platform to model platform on its own timeline, not in reaction to being out-featured.

Questions to Consider

• Can you state the single reasoning thesis behind your product's current roadmap in one sentence, and would your last three feature decisions actually trace back to it if you checked?
• Is there a metric-driven "standard trigger" (a headcount hire, a revenue milestone, a go-to-market motion) your team is following by convention rather than by evidence that it's actually where the value or margin sits for your specific market?
• When a competitor made news most recently, did your team's response come from your own product thesis, or was it a reactive scramble to match the headline?
• Is there a feature or workflow your product is best known for today that you would be reluctant to deprecate even if the market had clearly moved past it?

Bottom Line

The advantage that let a small team out-execute Microsoft, Windsurf, Cognition, and Anthropic's own coding tools wasn't a better model. It was refusing to hedge: one clear thesis about where value would accrue, applied with the same discipline to product, hiring, sales timing, and M&A, and the willingness to retire what made the company famous before being forced to.