Give the Agent a Way to Verify Its Own Work

The single most-repeated piece of advice from Boris Cherny and the Claude Code team: “Give Claude a way to verify its output. Once you do that, Claude will iterate until the result is great.” Without a verification loop (tests, type checks, a browser it can drive, a command whose exit code means success), the human is the only feedback signal, and every iteration has to pass through them. With one, the agent self-corrects until the work actually runs. Boris pegs this single move at a 2-3x quality improvement.

The same logic explains the guide’s companion rule: never let the agent claim success without evidence such as test output, screenshots, or real command results. An agent that cannot verify is guessing; an agent that can verify is iterating.

The deeper principle generalizes beyond coding: the quality of delegated work tracks the quality of the feedback loop you build around it, not the raw skill of the worker.