Before I get to my actual point, a little background: The current LLM based "AI" industry is effectively a stack of cards with services sold at a loss to promote dependency. A lot of the companies are just barely backed by incredibly circular debt agreements. Unless someone actually manages to create the sort of self improving AI psuedo-god they all claim is coming, it is literally not financially possible for the current environment to continue forever.

That means that eventually, prices will catch up and the likely free or cheap LLM functionality can and will be suddenly ripped away. It will also force companies to eventually examine how much value they are actually getting out of AI vs the eventually enforced real sustainable costs.

It's also worth noting that the current burst of AI tech (mostly just LLMs) is still very new. There's an argument to be made that we're in the dotcom boom, where everyone is convinced it's the wave of the future but very few have concrete plans about how and where to best use it.

So even ignoring the apocalyptic speedrun of killing the planet to try and squeeze better performance out, we can reasonably assume that a full AI future with countless jobs displaced is not likely. Prices will eventually rise, free money printers will eventually jam, and eventually it will start to be seen as a tool rather than magic.

It is my personal opinion that this is the best case scenario that we can find reasonable evidence pointing towards. Maybe we all die from climate change first, but at that point you'll have very different concerns.

Anyway, with that out of the way, I'll get to my point after the line

I'm going to go a little against the grain here. Until the inevitable bust happens you likely will be expected to use AI to some degree in your job. Especially if you're looking to do software engineering.

With that in mind, it makes sense to try and learn how to do it well. You don't need to cut it out cold turkey, but you need to pull back and try to do things the hard way more often, and you need to stop and take time to think critically about how and when you do use it.

Don't use it casually. Don't use instead of more traditional web searching. Use it as a tool to search in natural language if you don't know the specific terminology, and when it points you at a good source of information, you start building up your own knowledgebase.

Take your own notes. It's been proven over and over again that the process of taking notes helps to cement the knowledge in your head.

Track your time. Deadlines have and will be present, always. Sometimes you won't hit them. But if you practice planning you will get better at it over time. Then you can start planning how long you can afford to spin your wheels with any specific approach before you try a different one. One of those approaches can be AI, but it should never be your first one, and you should always strive to work through the problem and results yourself to ensure you fully understand it.

Any and all skills are like muscles. You can't just wake up one morning and bike the tour de france. You practice. You'll probably suck at first, but you can find ways to improve after each time. Eventually you'll get better. Some skills won't leave you even without regular practice, some will. But you're never just completely incapable of doing something just because it's hard to do now.

And some rambling asides: I dropped out of a Comp Sci degree, eventually found my way back to an IT help desk job, and worked my way up into Systems Admin/Engineering with a heavy focus on scripting automations and tools.

I had an almost five year gap between the last time I programmed and when I first started scripting things to make my help desk work easier. I remembered basics but I had to relearn a lot. I'm still relearning certain things like good organization of projects, properly psuedocoding and planning before starting to type out code, and splitting things into functions.

But it is possible. And I do occasionally use AI, but mostly to help point me at things I need to learn to fill gaps, or for assistance on the shape of a general approach to a problem I haven't dealt with before (dependency resolution can be hell, and Microsoft doesn't always make certain base level functionality with checking remote disk space straightforward).