I know it’s tiring to talk about “hallucination”, but truly, models still do hallucinate They constantly say they did a thing they didn’t, say they know how to

I know it’s tiring to talk about “hallucination”, but truly, models still do hallucinate They constantly say they did a thing they didn’t, say they know how to solve something when they don’t, etc. Regardless of guard rails or tests - AI forces a constant vigilance of a new kind. Not just “what might have gone wrong” but also “what do I think is working but isn’t actually”. And we’re not even talking about how it chooses substandard solutions, is happy to muddy code/architectures, add spaghetti on top of spaghetti etc. Agentic coding often feels like an army of unexperienced developers who are also incredibly eager to please.