in which we use AI even less? yes. dependency is RISKY

I never used AI for art or writing, except when Dalle-2 first came out (this was well before anyone outside of graduate level computer scientists even had access to it) where I generated like 1-2 images and went "nice, anyway, back to CtrlPaint". I've always hated AI-generated images, minus that first exploratory period, once the techbros got wind and tried to get rid of artists. No. Screw that. Lol. 

I used it for coding sometimes, but then realized it was making me weaker. I always refused "vibe coding". That term sounds extremely stupid because it is, in fact, an extremely stupid practice. Instead, I tried to use AI while understanding what it was doing and being the primary driver. But then I decided to try and go "AI-minimal" on code. I only used it if I was stuck on a physics formula or something that was difficult to source. 

But at work, I was still using it too much. Sure, we have copilot licenses, but how much was I actually learning? We can use it to speed up, but was I actually speeding up? Half the time I'd end up down bizarre rabbit holes; the other half of the time it would implement the feature cleanly and quickly. So, not necessarily, but sometimes. 

I just spent way too much time on a rabbit hole I didn't even need to go down as a result of a too-close-to-vibes generation. And this is with me being the lightest dev on AI use in my team. The token usage shows this. I started to realize that even my agentic approach was too flimsy. 

So I am once again trying an AI-minimal approach: 

• At home, I can use AI for things like: helping me find which articles to use, or asking a conceptual question that docs aren't explaining very well. That is IT. No AI code is allowed. 
• At work, I need to drive. AI must be used as a brake-check of sorts. I take control. I implement the features. I use AI to explain stuff to me if I'm stuck, and I use it to code review me, before the team does (that's policy, btw, so I have to do it). 
• if AI is trying to implement something confusing, I do not ship it. I ask if it's actually in-scope for the ticket. If not, nuke it. 
• If the AI cheerfully finishes my unit test for me or something (as in, autocomplete), I can look closely at it to make sure I understand every single line, and if so? Then ship it. But if there's anything I do not understand, no matter how small, it does not count!
• I do not have Copilot installed on any of my personal devices. Work ONLY. 

I also was using the chat-side too much. The chatbot side is not as bad -- I wasn't using c.ai or thinking ChatGPT was my waifu lol -- but I kept asking it random shit and then falling down exploratory rabbit holes that weren't a good use of my time. It was also very easily reinforcing my OCD. Asking compulsions and chatbots don't play well. 

While I'm less concerned with this than having my programming ability go down the shitter, I'm still trying to: 

• solve all problems myself first, especially if it's one where I most certainly can fix it myself, and am just getting lazy
• stop "spinning" on problems, as this is usually unpleasant 

the brain's dopamine reward system can be hijacked, not necessarily because of "sycophancy", because even if that's removed it's easy to spin. instead, you can hyperfixate almost eternally because the only chatbot that will get "bored" of your conversation is probably Gemini, and only hitting a context window overload will stop you. 

ultimately? 

• I don't need AI for art. 
• I don't need AI for writing. 
• I don't need AI for coding, except for code reviews at work only, autocomplete at work only, explaining thorny codebases at work only, and helping me with thorny physics equations outside of work (because let's be real, most existing explanations in physics-land suck). 
• I don't really need AI for chatting, but it can work as an advanced google or the occasional "why are people like this" -- but it's uncomfortably easy to fall into the "spin abyss", so this should not be done often (Kind of like eating molten lava cakes, which should also not be done very often because they are very delicious and will steal your sixpack). 
• I may use it for getting stuck on Strands (better than looking up the solution) or explaining to me some math concept, because every once in a while, even Salman Khan gets vague. 

Ironically I'm currently fighting the temptation to paste this very post into an LLM and ask it to provide an analysis. I don't need this. It's like a freaking gacha, even though I've never spent a cent on Genshin or Livly Island despite playing both regularly. I guess this is what it's like. 

Compulsive. 

I think I need to work on Zig so that way I won't be allowed to use it at all. 

Not trying to be dark, but: Why do you think these companies are pushing AI on you so hard? Not to tinfoil hat, but maybe they want you to...be dependent on their products. If you lose your edge, you lose everything. 

What will you do when companies ask you about AI use, clearly wanting use of it? Well, I'm not someone who is never going to touch it (at work, because it is unrealistic), so I will answer honestly: It can speed up your workflows. You just need to be the driver and own all your code. You are the one responsible for it, not Copilot. 

Purists won't like this, but purists, unfortunately, are almost never SWEs. AI can be like Kite autocomplete (RIP) but better. Or, it can be a way to produce slop software. Garbage. I hate the concept of vibe coding because a) it sounds like a meme, and b) by definition it implies that you are actually aiming to understand as little as possible about the underlying code. That's just gross. It's a symptom of our modern USA society in general, though. Shitting on education. Shitting on learning. No one cares anymore. 

So, slop or assisted learning. 

Which will you choose? 

Also, 

> implying I'm going to apply for jobs when I like my current one 
HAHAHA 

no

not until I can't take winter anymore I ain't