Agentic Coding

207: Work Without a Test Suite

How do we verify work that no test can run? Much of what an agent produces is not code. It drafts documentation and reader-facing prose, and it rewrites its own working records each time it folds them for the next session. None of that work has a suite to go red. This chapter teaches three checks for it: a gate for prose, made of a mechanical pass, a hostile read answered in a ledger, and a strip of process residue; an audit that reads a document as a set of claims against the as-built code; and a fold checked for what it dropped and for what it wrote.

The chapter closes the second year. The third year begins with 301: The Step Up, where one human runs several agents at once.

Materials

This Chapter's Insights

Browse every insight on the course map.

Downloads