Reflecting on a Year of Claude Code (Boris Cherny)
- Source
- youtube.com
- Author
- Claude
- Date
Context
An checking its own work needs a way to exercise the finished feature. In Anthropic's Claude Code retrospective, a team member describes a desktop-development that launches the local application and lets Claude click through a new interface, try edge cases, fix problems and check again.
The procedure itself also gets maintained. When the agent runs into a staging-environment problem and debugs it, the speaker has it update the skill. That turns a one-off debugging session into instructions available for subsequent work. Verification here means running and observing the application, beyond the unit tests, linting and type checks a project usually automates already.
- 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.
- agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
“I tell it to write it to the CLAUDE.md, or to like make a skill or or something to do it differently. And if you can do this, then Claude can just like run forever.”
Boris Cherny
“It's just human nature, when you accept 99% of requests, that your eyes just glaze over when you read it.”
“I don't talk to an agent anymore. I talk to loop or I talk to a routine and it prompts Claude for me.”
Boris Cherny
“And the thing that I would not have expected six months ago is probably half my engineering now I do on my phone.”
Boris Cherny
“You give it the minimal possible system prompt, the minimal possible tools, and then you let the model figure it out.”
Checking sign-in…
Loading comments…