The AI integration is interesting to me. I have been trying to get Claude to help me with postgres from within my product and have found its ability and understanding of postgres/sql to be significantly worse than with common programming languages.

Maybe it's because my tolerance for imperfection is much lower for databases than web apps, but it is so bad that I can't trust it for anything database-related beyond text transformations and generating boilerplate queries. It will incorrectly tell me things like postgres doesn't support TABLE OF <TYPE>, or make syntax errors for ON CONFLICT, and immediately agree with any "what about" or "are you sure" I throw at it.

Curious if anybody else has run into this. Obviously LLMs are not always great in specialized domains like this but the poor performance with something as popular as postgres is pretty uncharacteristic IMMO.