The Creator of Claude Code on The Hottest Piece of Software in the World
Source
omny.fm
Date
Key takeaways · AI-distilled
Cherny's case for why a safety lab built a coding tool: a model has no body, so the way it acts on the world is by writing code. Getting good at code is therefore inseparable from making the model safe to act.
evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → and interpretability inspect a model in the lab, but Anthropic's position is that you only learn whether it is safe by shipping it and watching people do real work with it. The product doubles as a safety instrument.
Claude 3.5 was the break point: models went from finishing a line after a few keystrokes to writing a whole file or feature. That capability jump, not a product roadmap, is what made an agentic coding tool worth building.
Terms in this piece · Glossary
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Why it matters
The engineer behind Claude Code explains how the fastest-growing coding AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → is built and where agentic development is heading — first-hand detail you can't get from release notes.
Key quotes
“Our computer language is going to be irrelevant in the future.”
“Everyone moves up to the higher order, more abstract thinking.”
“We're getting better at code, we're getting better at writing.”
