LLMs are great for problem exploration. Especially with the decline of Google I think we're at a point where it's less difficult to get an LLM to spit out something that'll accomplish a task sorta compared to actually finding that on the internet. But if the task is going to be repeated or modified then I think LLMs are at a permanent disadvantage to prebuilt software. Even if that prebuilt software is just someone else running an LLM and then passing the output through acceptance testing most people just don't want the headache of debugging weird edge cases and the novelty of "I'm a developer too!" wears off pretty quickly.

I'm excited that the weird grey-zone of excel sheets with business critical logic is likely going to disappear as LLMs slowly make the logic driving those too complex to be comprehended and managed and those get foisted off onto actual engineering resources. It'll be painful but probably for the best... but for actual tools people need to use day-to-day I think the assurance that the tool will work has a lot more value than the AI hype has comprehended.