We combat this with, funny enough, AI code review.

Much MUCH stricter structural, consistency, quality, and discovery standards on code through rules and docs. Well defined expectations, stricter testing standards…etc Then let some agents loose on PR review and they catch all sorts of problems. Especially with other AI slop.

Coincidentally, the process of doing all this actually reduces the sloppiness of the slop that we receive, as the AI tools writing this code consume these rules before generating code.

Sad state of afairs. Many awesome projects will get tainted into history.