I'm not talking about GH Copilot or some autocomplete IDE feature, I'm talking about fully autonomous agents 2-3 years in the future. Just look at the insane rate of progress in the past 2 years. The next two years will be even faster if the past few months are anything to go by.
Are we living on the same planet? I haven't seen much real progress since the release of chatGPT. Sure brenchmarks and graphs are going up, but in practice, meh...
I am completely blind and I used Gemini Live mode to help me change BIOS settings and reinstall Windows when the installer didn't pick up my USB soundcard. I spoke, with my own voice and completed a task with a computer which could only see my webcam stream. This, to me, is a heck of a lot more than ChatGPT was ever able to do in November 2022.
If you continue to insist that stuff isn't improving, well, you can in fact do that... But I don't know how much I can trust you in terms of overall situational awareness if you really don't think any improvements have been made at all in the previous two years of massive investment.
For the same reason you need Engineers to operate CAD software.
You can outsource the execution of a task but only if you know how to formulate your requirements and analyze the situation.
What fundamental limitation in AI makes you believe reasoning agents in 2 years won't be able to do this?
AIs have been shown to have confirmation bias depending on what you ask them. They won't question your assumptions on non-obvious subjects. Like "why should this application be written in Python and not C++". You could ask it the opposite, and it will provide ample justification for either position.
What fundamental limitation in AI makes you believe reasoning agents won't be able to gather requirements, ask clarifying questions, and make decisions in 2 years?