Maybe they are speeding up things that were not the bottleneck or don't matter. Or LLMs speed up the time to learn, "well this idea probably sucks actually", which is great, but in the end there is still no good idea. For software dev, it had already gotten pretty efficient before LLMs came around - the business of making software demanded it. There are still gains to be had no doubt, but many will have less impact.

I'm using claude to make gaming mod apps I never would have bothered to code before, but they have zero monetizable value. There is demand for sure, but somewhat ironically, copyright/IP/TOS law stands in the way. The thing which committed the biggest copyright infringement in human history lets me make another thing which I can't release because of copyright law. I'd get cease-and-desisted instantly if I did so.

The latest BCG report I saw on this was the one from this summer that found that employees are saying they're saving time, but businesses still haven't figured out how to turn that into dollars.

And employees self-reporting that they saved time doesn't necessarily mean they actually save time. Humans are famously crap at estimating that sort of thing.

But does crazy speedup of tasks actually yield more productivity? Or does it yield greater numbers of pointless tasks?

Anecdotally, I've seen management using AI to produce more and more flashy looking dashboards and interactive displays of various metrics. Will this help our bottom line?