You update a prompt. You test it on five examples. Looks good. You ship.

Three days later, a support ticket: “$4.2M in Q3 revenue” → “strong revenue growth”.

Nothing crashed. No alerts. Just a silent regression.

With code, diffs exist automatically. With AI, they don’t – unless you compute them.

I built Verist, an open-source library that brings code-style change control to AI decisions: replaying past inputs and showing exactly what changed when you tweak a prompt or model.

First public release (v0.1.0) is out. Full write-up and links in the comments 👇

#OpenSource #AIEngineering #LLM #SoftwareEngineering #DeveloperTools