Developer Uses Timestamped Code Comments to Track AI Collaboration and Prevent Regressions

A software developer building multiple AI-assisted apps found that code comments rarely captured the reasoning, confidence levels, or failed experiments behind decisions — only what the code did, not why. To address this, they created a lightweight convention called MurphySig, embedding structured comments that log which human and AI model wrote the code, the date, a confidence score, and any follow-up review notes. Across 14 repositories and roughly 450 signed files, the practice helped prevent a real regression when a later AI model nearly re-introduced a bug that an earlier signed note had already documented and explained. Benchmarking showed that providing explicit context in any form — not just the MurphySig format — reduced AI fabrication of code authorship from 11% to 0% and improved model briefings. The author argues the same principle applies to AI agents whose identity files drift over time with no record of who changed what, when, or why.