They present me elaborate architectural documents, troubleshooting documentation, helper scripts, etc - and it all just doesn't really fit together. And those people don't have the slightest clue what they're doing, besides letting their agent cobble everything together, seemingly working until it doesn't - and then I'll have to deal with it.

I'm thinking about putting up some limits here - allowing myself to reject those cases as soon as it turns out that there was #AI involved.