Self-reinforcing quality / adopting verification pattern

Imprint's engineering team makes every AI-written pull request carry a file listing what the AI agent checked and what it could not check, Will Larson writes.

Will Larson is an engineering leader at Imprint, a financial technology company that runs co-branded credit card programmes. He found that pull requests from AI agents often reached the shared build system failing basic checks that could have run on the developer's machine. His team now requires a written verification file in every pull request, and reviews those files each day to improve how the AI agents work.

• Each pull request must include a file listing every step the AI agent took to verify the change, and what it was unable to verify.
• A pull request cannot move from draft to ready for review until that file exists. An AI review tool comments on every pull request, and weak comments are dismissed with a stated reason.
• Every day the team updates the AI agents' guidance from where verification failed, opens issues for the gaps, and checks which review comments were accepted or rejected.
• Larson reports that each major code area, namely frontend, backend, data and native apps, starts the day with four to six pull requests from this loop to review.

A team that already reviews AI agents' pull requests can add the verification file without new tools, because it only asks the AI agent to write down its checks. This is an inference and is not the author's claim. Larson notes that some colleagues doubt how far Imprint should centralise this guidance.

Derived by Working Surface from the article; more in the Playbook. Source line: Now the harness has to explicitly document that it’s done those steps, or acknowledge it hasn’t.

31 October 2026: as AI agents produce more of the work, teams move the checking to the people and records closest to the real output.