Boris Cherny

Boris Cherny created Claude Code at Anthropic, where it began in September 2024 as a terminal chat built to test the API. He tells the origin story the same way in several places: no interface to build, so a CLI; then Sonnet 3.5 spontaneously wrote an AppleScript to identify the music playing, and the read was that the model simply wants to use tools. What followed was six months where the bet on having the agent write all the code did not work, a first release that was not a hit, then Opus 4 in May 2025 as the inflection point.

The reason he keeps recurring is that he reports on himself as a test case. He says he has not written a line of code in 2026, that the model has written 100% of his code since October 2025, and that his record is 150 PRs in a single day. He runs five agents in parallel, works mostly from his phone, and uses «Opus 4.5» specifically because the slower model needs fewer corrections. His «conseils équipe Claude Code» push git worktrees, plan mode, «settings.json», and «subagents spécialisés (code-simplifier, verify-app)».

He also supplies the framing others reuse. "Loops are the future" and his shift from prompting to writing loops became the seed of «Loop Engineering» in Andrew Ng's and Shubham Saboo's hands. His "Steps of AI Adoption" ladder puts Anthropic at step 3 and himself at 4, and argues that Anthropic's edge is organizational, not technological. What that ladder does not settle is who says no once everyone can build.