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.
Correct framing.
Many here are idiotic. Economic growth is measured via monetisation of output - not what you think is of value to you.
For my part I’ve seen it actually slowing down time to decide ideas suck. It’s reduced the friction to start implementing so much that people are just diving in before they’ve even had a chance to fully bake the idea. At which point they’re likely to get stuck on Mr Sunk Cost’s Wild Ride.
Management doesn’t see the problem because they are still completely distracted trying to measure dev team productivity in story points or KLOC instead of something meaningful like USD.