https://blog.stuartspence.ca/2025-08-declining-ai-slop-mr.html some useful criteria for review thresholds for slop PRs

Yup, a lot of this tracks well 👍🏻

Which is, however, partly the fault of very dogmatic linters / static type checkers. They will, unless configured otherwise, based on the inferred types flag, if an edge case was not handled, if it is technically possible. And the agent will continue with yet another iteration of the code, until no more errors occur. I have that currently a lot with the new freezer database that I am building for NGI. For every

try

, I have to specifically look up and import the error types that I want to

catch

, because I can no longer simply catch

any

exceptions.

Sometimes I think it's better to let some exceptions go unhandled 😅