[flagged]
What are you using to measure 'better' here? And what professions come to mind when you wrote 'almost all professions'?
If I think about some of the professionals/workers I've interacted with in the last month, yes they _could_ use an LLM for a very small subset of what they actually do, but the error it can introduce if relied upon (the 'average' person is implied to not do their job as well so relying on the output is likely now or into the future?) I would wager makes things worse in their current state.
It might get better, especially over such a long time horizon as 20 years, but I'm not expecting them to be recognizable as what we currently have with the SOTA LLMs (which is mostly what people are currently referring to when using the marketing term that is 'AI'). And in the long term focusing/relying on 'AI' to improve the ability of professionals in 'almost all professions' is IMO just the wrong thing to be sinking such a large amount of money into.
If you need to jump into large, unknown, codebases LLMs are pretty damn good at explaining how they work and where you can find stuff.
A lot faster than clicking through functions in an IDE or doing a desperate "find in project" for something.
And just sticking a stack trace in an LLM assistant, in my opinion, in about 90% of the cases I've encountered will either give you the correct fix immediately or at the very least point you to the correct place to fix things.