television_onesettings_one

Blog

← All posts

Are you still using Cursor: Future of AI Coding Platforms

Are you still using Cursor: Future of AI Coding Platforms

Are you still using Cursor? Here’s the part most startups are missing.

Cursor is a monster success on paper: it’s reportedly used by 67% of the Fortune 500 and generates ~150 million lines of enterprise code per day. That kind of distribution is what every startup founder dreams about. But in my opinion, the real story isn’t “AI coding is here.” It’s where the power sits in the stack. In my experience, the biggest pitfall for an AI coding platform isn’t building a great editor experience—it’s platform dependency. Cursor’s core risk (per the reporting) is that it’s built on models controlled by others, while those same model providers (OpenAI, Anthropic) are also building their own coding tools. That’s the classic supplier-competitor squeeze.

One concrete tradeoff I’d plan for (and I’ve seen teams underestimate)

Stay a thin product layer on top of frontier models:

  • Pro: you ship faster and ride the model curve.
  • Con: you’re exposed to bundling, margin pressure, and feature copycats from the model owners. Try to build your own models:
  • Pro: more control, less dependency.
  • Con: massive cost and focus drain (capital + talent), and you may still lose on distribution. BTW, what fails for a lot of startups here is betting that “better UX” alone will protect them. If the underlying model provider can bundle “good enough” coding into their ecosystem, your differentiation gets thinner than you planned.

Where I think the durable moat shifts next

As AI gets embedded in daily work, EQ becomes a rare human advantage—things like building trust, navigating tension, and making people feel seen (the stuff AI still struggles with). The next generation of coding platforms won’t just understand code; they’ll need to understand the human on the other side: frustration, uncertainty, confidence, and when to slow down vs push forward. If you’re building in this space: what’s your plan to escape the API dependency trap—and what’s one “human understanding” feature you’d bet on that model providers won’t prioritize?