Correct me if I'm wrong, but I think this argument hinges on AI not really being intelligent.

If we're talking automation in the way that humans define a process, boundary cases, etc up front then yes you can't really automate away one-off monotony.

If AI is actually artificial intelligence, it will be able to figure that out similar to how a human would.

▲

sublinear 4 hours ago | parent [-]

I compared it to interrogation because it's about steering the conversation. The only reason you're talking to middle management is because their work is getting outsourced to SaaS.

If you waste too much time on irrelevant topics, you lose. If you don't collect enough information, you lose. If the middle manager feels threatened, you lose. If you don't have a good production release, you lose. If there are any high severity bugs, you lose. If there are too many bugs, you lose.

The only way to win is to carve an exact path through the mess from the beginning, and that needs human experience.

	▲	_heimdall 4 hours ago \| parent [-]
		Maybe we just have different ideas of what AI would be? I don't consider an impressive text predictor to be artificial intelligence. Maybe that means LLMs aren't AI, I honestly don't know because no one seems to care about understanding how they work or solving interpetability first. I do expect anything that earns the banner of AI could ask good questions, weed through a bunch of word vomit to find the key nuggets, and act on them similar to or better than a human could. Anything less than that doesn't seem particularly intelligent.