AI rewrites pass tests but fail the trust question

A Postgres rewrite in Rust is passing 100% of the Postgres regression test suite, and the HN thread is not celebrating. The top comments make the same point independently: Postgres's reliability comes from decades of production hardening, edge cases discovered in anger, and judgment calls baked into the codebase. Tests capture behavior, not wisdom. One commenter draws the direct comparison to SQLite, which is also not test-reliable because of tests but because of a different kind of rigor entirely.

There's a secondary debate about the AGPL license choice. Postgres itself chose MIT-style licensing and built a sustainable ecosystem without copyleft. Choosing AGPL for a Postgres rewrite reads to some commenters as a defensive move, or possibly a commercial one, which changes the trust calculus for anyone considering adoption.

The broader pattern: several threads today are implicitly asking the same question about AI-generated or AI-assisted work. The train sim thread asks 'how much was LLM involved?' The Bun/Zig rewrite debate is happening in parallel. The community is developing a reflex to ask not just 'does it work?' but 'how do we know it works, and who vouches for it?'