Boris Cherny’s latest post on X argues that engineering teams should treat automation and documentation as the main way to make agentic coding systems more useful, not just faster. In the message, Cherny writes that the best engineers have long spent time on “better vim/emacs automations,” lint rules, and e2e tests, and he claims those habits matter even more now that agents can speed up routine work.
Cherny’s point hinges on moving recurring work into code. He writes that if an agent fixes an issue one time, it may still spend tokens and miss edge cases, while a lint rule, CI step, or routine can automate that class of problem “forever.” He also argues that automation helps more people contribute to a codebase, including people who are not familiar with it, because the knowledge no longer lives only in someone’s head.
He goes further, suggesting that teams should write “CLAUDE.md,” “REVIEW.md,” skills, and docs that let agents work in a codebase with “zero additional context from the prompter.” A companion graphic attached to the post distills the idea into a slogan: “THE REPO SHOULD TEACH CLAUDE HOW YOUR TEAM WORKS,” alongside the line “Domain knowledge → infrastructure → day-one contributions.”
The post drew a flurry of replies focused on standards and documentation files, with some commenters backing the idea of shared guidance files such as AGENTS.md. Others pushed back, arguing that guidance files are often ignored or that relying on markdown for automation can become unwieldy. Cherny’s post itself, though, sticks to a familiar engineering argument: encode repeat knowledge into infrastructure so fewer decisions have to remain manual.
Source: X
