The Creator of Claude Code Says He Hasn't Written Code in 2026

Boris Cherny, the person who built Claude Code, told Sequoia's AI Ascent that he hasn't personally written a line of code in 2026. He delegates 100% to AI. Ships dozens of PRs daily from his phone. Predicts the title "software engineer" starts disappearing by end of 2026, replaced by "builder."

His exact framing: "Coding is solved."

I use Claude Code every day in my personal projects. I've shipped three products with it in the past year. I'm about as bullish on AI-assisted development as anyone. And I think Cherny is both right and dangerously wrong at the same time.

He's right that the mechanical act of writing code is increasingly automated. I spend more time reviewing, directing, and evaluating than typing. The bottleneck has moved from "can I write this function" to "should this function exist" and "is this the right abstraction." That shift is real. I feel it every day.

But "coding is solved" collides with DELEGATE-52's finding that AI silently corrupts 25% of content in long workflows. It collides with 380,000 vibe-coded apps shipping with no security. It collides with the reality that Cherny is an exceptional engineer who can evaluate AI output at a level most people can't. When he reviews a PR from his phone, he brings 15+ years of context about what good code looks like. Strip that away, and "coding is solved" becomes "generating plausible code is solved, but verifying it is now the entire job."

He also said every team member at Claude Code, including PMs, designers, and finance, now writes code. And he believes Claude Code itself may be 100 lines a year from now. That second prediction is the more interesting one. If the tool itself becomes radically simpler, the skill ceiling drops and the "taste" ceiling rises.

Andrew Ng seems to agree on the direction if not the speed. His new free course with JetBrains on spec-driven development explicitly addresses the gap: vibe coding is fast but unreliable. The answer isn't "stop using AI." It's "get better at specifying what you want and verifying what you got."