Posting here because a lot of what breaks in AI-built apps shows up right after a Vercel deploy — works locally, ships, then something’s subtly wrong in prod.
I built Khwand to catch that earlier. It’s a GitHub App that runs on every push: extracts your Python functions via AST, generates adversarial edge-case tests specific to what the code claims to do, runs them in sandboxes, and opens an auto-fix PR if something fails. Sits before your Vercel deploy step, not instead of it.
Built it after one too many “AI agent shipped it, tests passed, prod broke days later” stories — the gap is that standard tests assume human-written code with human assumptions, and AI agents don’t share those assumptions.
Python-first right now. Free during early access: khwand.com
Curious if others here are seeing the same pattern — AI-shipped code passing local checks and breaking specifically after deploy.