Claude Code Creator Left for Cursor, Came Back

Boris Cherny, the creator of Claude Code, left Anthropic for Cursor. Two weeks later, he came back. That's the most important data point in the AI coding tools war right now.

The story came out in a Lenny Rachitsky interview with Cherny, and it's been lighting up tech Twitter ever since. Not because job-hopping is unusual in Silicon Valley — it's practically a sport. But because the person who built the thing chose to come back to it, after seeing the competition from the inside.

The Numbers Are Absurd

Let's talk about what Claude Code has become while everyone was busy debating whether it was "just a wrapper." The tool now accounts for 4% of all commits on GitHub. Four percent. Of the entire platform. Let that sink in.

Daily active users doubled last month. Not grew 20%. Not had a nice quarter. Doubled.

For context, GitHub has over 100 million developers. Claude Code isn't some niche dev tool anymore — it's becoming infrastructure. The kind of thing that, if it went down for a day, would measurably slow the output of the global software industry.

Why He Left (and Why He Came Back)

Cursor is, by most accounts, excellent. It's the darling of the AI coding tool space — fast, polished, deeply integrated into VS Code. When Cherny left for Cursor, it was a reasonable move. Maybe even the obvious one. Cursor had momentum, a passionate user base, and the kind of product-market fit that makes investors salivate.

But two weeks in, he came back to Anthropic.

We don't know exactly what he saw at Cursor that made him reverse course. But we can make educated guesses. Claude Code operates at a fundamentally different level — it's not an autocomplete engine bolted onto an editor. It's an autonomous coding agent that can reason about entire codebases, plan multi-step implementations, and execute them with minimal hand-holding.

When you've built something that writes 4% of GitHub and you see the competition up close, one of two things happens: you get excited about the new opportunity, or you realize what you already had was better. Cherny chose door number two.

The Claude Code vs. Cursor Narrative

This story matters because it crystallizes the real split in AI coding tools. It's not about features or pricing anymore. It's about philosophy.

Cursor's bet: AI should augment the developer experience inside the editor. Better autocomplete, inline chat, smart edits. The human stays in the driver's seat, and the AI is a very good copilot.

Claude Code's bet: AI should be a peer programmer that can take on entire tasks. You describe what you want, and the agent figures out how to do it — reading files, writing code, running tests, iterating. The human becomes the product manager.

Both approaches work. But they're aimed at different futures. Cursor's future is one where developers are 3-5x more productive. Claude Code's future is one where the definition of "developer" changes entirely.

What This Means for the Market

The AI coding tools market is consolidating around these two poles faster than anyone expected. On one side, you have the editor-integrated tools: Cursor, GitHub Copilot, Windsurf (née Codeium). On the other, you have the autonomous agents: Claude Code, Devin, Codex.

The editor tools have the user base advantage. Developers already live in VS Code. The agent tools have the ambition advantage. They're not trying to make today's workflow faster — they're trying to replace it.

Claude Code's growth numbers suggest the agent approach is winning converts fast. When a tool doubles its DAU in a single month, it's not because of marketing. It's because people try it, have their minds blown, and tell everyone they know.

And the 4% GitHub stat is the kind of number that makes enterprise CTOs pay attention. If Claude Code is already writing that much code, the question isn't whether to adopt it — it's how fast.

The Most Telling Signal

Forget the metrics for a second. The most important thing about this story is that the creator came back.

Boris Cherny had the ultimate A/B test. He built Claude Code at Anthropic, then went to see how the other side lived at Cursor. He had full context on both products — not from demos or blog posts, but from actually building them. And he chose Claude Code.

In a market flooded with hype, benchmarks, and cherry-picked demos, that's the clearest signal you can get. The person with the most information made their choice. And they made it fast — two weeks fast.

The AI coding tools war isn't over. But the Boris Cherny boomerang tells us something important: when you strip away the noise and let someone with deep expertise evaluate both approaches, the autonomous agent wins.

That's not a marketing claim. That's a revealed preference.

Follow ultrathink.ai for daily AI analysis that cuts through the noise. Subscribe to our newsletter for the stories that matter.