I’m not in a computer science or software engineering field. I have been coding for a long time, started with turbo pascal, if that tells you something. AI (LLMs) is horrible for many things, especially writing things for humans, or generating “answers” for people’s questions about the real world.
AI agents are very good at writing code. If a human needs code for work, they are better off telling a well configured and prompted AI what they want done and how, than writing code themselves. There may be certain exceptions, mostly at the more sophisticated levels of performance-critical systems, but even then, AI will make the end product better by helping the human developer by writing and running tests, testing alternative approaches, etc. , because with AI assistance, the dev can simply get more of that done within the given timeframe.
That is all to say, I can’t imagine a job that involves writing code that doesn’t massively benefit from AI. It’s a tool that should be used. You should learn how to use it well, it’s part of your trade. Like you don’t hand punch cards to run machine code. You use an IDE, code completion, version control, linter, etc. You learn these tools as part of your education, because if you don’t use them well, you’re not good at your job. Of course if you code for hobby or whatever, this doesn’t apply.
The comments about after the crash… Agents for writing code will stay after the crash, like the internet stayed after the dotcom bust. Like rn I pay 20 bucks a month for claude, for work. That’s unsustainable, it will and it should cost more. If it costed $2000 a month, I would still keep the sub, without any doubt. Just for comparison, 3 years ago I got a quote from a data science consulting company to maintain a specific project for 30k a year- that didn’t include developing it. A single Claude pro sub is objectively better for us than what that contract offered.