Trend I’m noticing: Live LLM demos often fail or provide invalid/surprising output. “Oh weird, this worked fine in my testing” If I can’t ship with confidence, that’s a big problem.

@housecor That will continue for a very long time because the thing doesn't know what it is doing. It always gets something tiny wrong. It is mainly good for getting boilerplate for now. The speed got from AI is like taking loans from your future self but to be paid with interest.