I think that 20 years from now, we'll all be sitting around wondering 1) where the fuck are my flying cars, and 2) what were they thinking using computers to write code?

And the reason I say this is because these tools are answering a question that we haven't asked yet: what common problems need to be solved in this programming language, and where do I get code to solve that problem?

These LLM modules are basically telling us how to duplicate code, and what we need is the opposite: how to stop reinventing the wheel for the 100th time.

Instead of writing code for me, tell me if I already have it. If I'm writing it, tell me there's a library for that. If I'm a library writer, give me suggestions for what libraries are missing from the toolkit.

All we've done so far is begun the process of automating the production of duplicate code. With absolutely no way to go back in time and correct bugs introduced in earlier iterations. We are likely, for instance, to see 0 day attacks that affect hundreds of applications, but with no simple way to describe which applications are affected. That's going to be a first rate trainwreck.