Comment by ianbutler
3 days ago
Yeah I've had problems with this recently. "Oh those are just warnings." Yes but leaving them will make this codebase shit in short time.
I do use AI heavily so I resorted to actually turning on warnings as errors in the rust codebases I work in.
Easiest to have different agents or turns that set aside the top-level goal via hooks/skills/manual prompt/etc. Heuristically, a human will likely ignore a lot of warnings until they've wired up the core logic, then go back and re-evaluate, but we still have to apply steering to get that kind of higher-order cognitive pattern.
Product is still fairly beta, but in Sculptor[^1] we have an MCP that provides agent & human with suggestions along the lines of "the agent didn't actually integrate the new module" or "the agent didn't actually run the tests after writing them." It leads to some interesting observations & challenges - the agents still really like ignoring tool calls compared to human messages b/c they "know better" (and sometimes they do).
[^]: https://imbue.com/sculptor/