In February 2025, Claude Code wrote about 20% of Boris Cherny's code. By May, it was 30%. By November, it hit 100%. Cherny, who leads Claude Code development at Anthropic, now ships 10 to 30 pull requests daily without writing a single line himself.
Before Anthropic, Cherny was among Instagram's most productive engineers and led code quality across all of Meta. Now he spends his time orchestrating AI agents rather than typing code. In a recent blog post, he drew a parallel to an unlikely hobby: making miso in the Japanese countryside, where he lived before joining Anthropic. White miso takes three months to ferment. Red miso takes two to three years. You set the conditions and wait.
The analogy fits his current role more literally than most tech metaphors. Directing AI coding agents is closer to managing a fermentation process than to traditional programming. You define the environment, set constraints, and let the system do its work rather than forcing every outcome by hand.
The broader numbers back up the shift. Anthropic says its engineering team grew 4x while productivity per engineer increased 200%. Claude Code currently accounts for 4% of all GitHub commits, with a prediction of reaching 20% by end of year.
These are self-reported figures from the company selling the tool, so take them accordingly. But the personal trajectory is harder to dismiss. A former Meta engineering leader going from 80% manual coding to 0% in nine months is a concrete data point about how fast this transition is moving for people at the frontier.
When asked what he would do after AGI, Cherny said he would probably go back to making miso.