We don’t AGI for LLM to be useful.

And regrading comparison with self driving car they are still improving just the bar for them is much higher. If autopilot works 99.9% if time then 1 out of 1000 drivers will die - so technology has to be even better. for LLM is enough if it’s 90% good to be broadly useful.

Not to "replace programmers", though. For the same reason you can't replace drivers even though they also make mistakes.

It’s not about replacing all programmers. If one programmer with AI assistant can do work the same as 2 programmers then one position is redundant.

Even with self driving truck if one truck driver is leading another truck behind controlled by AI, and just for safety you have somewhere C&C center with one person monitoring 4 such ai trucks and in case unexpected event remotely take over control then one truck driver position is redundant.

While I do think there is some threshold where increased productivity makes positions redundant, I don't think 2x would do it in most orgs. My current team easily has enough work for us all to be 2x more productive.

unless you want the LLM to write the firmware of the self-driving car.

fwiw, self-driving cars did not grind to a halt, development just did not move as quickly as the pundits and self-promotion claimed. I just rode in a fully driverless car on public streets in downtown Austin this week.