[codingdave]

So based on the comments, I'm the only person who think our jobs will be diving into years-old LLM-generated code to fix it all, and laughing: "Oh my, do you all remember when the entire industry burned 5 years and b/trillions of dollars just to try to force a use case to work that wasn't even the right match for the tools?"

I'm not even trying to be snarky. I just see that LLMs already work fine in almost every other use case. Better than fine - they have been implemented and they work, and don't need much more advancement. But coders are still fighting it, coming up with more and more elaborate setups, spending more and more money, just to get to a level of coding that is passable. Not great, not better than human, just a "CRUD apps work." level of passable.

Once the exec teams stop pushing the FOMO from down on top, I think people will settle down into a much lower level of AI assistance. It will continue to be used, but we'll have it boilerplate the simple stuff out for us, while we focus on the meatier problems. And that is not all that different than pre-AI: whip the easy stuff out, then work on the hard stuff.

[Sheeny96]

How long ago did you use the tooling? I've seen this take a lot, and I can understand it if you're looking at the output of Sonnet 3.5 copy and pasted. But models like Opus 4.8, that just is not the case. Sure, you can't just whack it into the codebase, it needs review. But lets not pretend that 85% of programming jobs are CRUD with some async messaging. And the newer models are perfectly serviceable for that sort of work.

[csomar]

A parallel is crypto. There are thousands of multi-million dollars projects that are now completely abandoned. It's the web otherwise this thing will look like a Chinese ghost city. AI will have a similar fate except that it seems like they are about spend orders of magnitude more money.

AI is useful as a retrieval device but not for planning/writing foundational code. Maybe UI/UX in React but for most other things it's a time bomb.