We're talking about AI—well, actually LLMs, because I refuse to use that misleading "artificial intelligence" label.

Yes, I use LLMs for coding. So do my colleagues. So does the entire IT industry. GameDev too, though they still pretend it's a bit "meh" while sipping craft beer and debating evil billionaires flying private jets to the grocery store.

Anyway—I use LLMs for coding and I'm not embarrassed about it. I won't hide that when it comes to coding, I was somewhere in the lower half of the median. LLMs give me a handicap. They let me build things I could only dream of before. And I completely get it if someone chooses not to use AI—but in my opinion, that only makes sense if you're already in the top half of that median. If AI averages out code quality, then outstanding developers might actually experience a slight downgrade. For now. Because in a moment, that might become a problem for them too.

Here's what my LLM journey looks like:

I've been testing models for four months now. Currently spending around $80–100/month across various subscriptions—though that's recent; before that, I capped at $40/month. I'm probably overpaying a bit, but I have a firm rule: while things are unstable, I stick to monthly subscriptions and switch providers like gloves. Changes happen so fast that what was the leader a week ago might not be the next.

My current top picks:

1) GPT-5.5 xhigh (OpenAI) – Does all the heavy lifting in Unity. Handles everything I throw at it. Large context window, rarely makes mistakes. Ever since Claude started playing games with limits and models, I've mostly stuck with this—to the point where I exhausted my monthly limit in three weeks and had to buy extra tokens.

2) GLM-5.2 – Enter the Chinese saviors, stepping in while evil America tries to make us slaves. I use GLM-5.2 alongside GPT-5.5. Quality-wise, there's no difference. I've reached the point where sometimes I run both simultaneously—GPT on one machine, GLM on the other—while I read what the robots are cooking and handle documentation, paperwork, or learn non-coding things (like Blender—more on that later).

Key point: GLM is open-weight, so theoretically you could run it locally. In theory—because obviously no one's setting up a data center at home... BUT... it's as good as GPT or Claude and ALREADY does a fantastic job. Hardware will get cheaper soon. And even if it doesn't, smaller providers and alternative solutions will break the American oligopoly. And that makes me genuinely happy—I hate being on a leash held by techbros who couldn't care less about protein-based intelligence.

Way behind:

3) Gemini 3.5 Flash & Kimi 2.7 – Not bad at all, but compared to the leaders, not the same league for the same money. Gemini edges out Kimi slightly because Google throws in nice extras (e.g., 5TB of Google Drive space). Kimi also has a much smaller context window... BUT... if you're doing webdev, both are more than sufficient.

What I've built in just 2 months:

(Excluding paid work I can't discuss without approval)

  1. Nearly finished the entire vertical slice of Chains of Fury—just lots of manual fine-tuning left

  2. My own blog engine (no SQL), which I forked into three branches:

    • A gamedev & tech blog

    • A private opinions & capoeira reflections blog

    • The official RDA Polonia capoeira section website—expanding it for multiple sections, editors, with learning content exclusive to RDA students

  3. A calorie tracking & diet monitoring tool

  4. A browser game in Three.js based on Pandora's Box

And that's just 2–3 months. I'm deliberately not dropping links—each project will get its own post.

What I DON'T use LLMs/AI for—and won't:

Graphics and content creation.

I refuse to hand over my creativity in game design to AI.

Why?

  • Generative graphics are, well... generative. AI handles styles that are—not terrible, but uninspired. And imagine everyone having access to the same tools. If you want your game to stand out visually, AI won't get you there.

  • If I don't need to grow further in coding (my skills are "good enough" to get things done), I want to develop other abilities—ones that make me more self-sufficient. Besides, creating is fun. It gives me that dopamine hit. I'm not giving that up to a machine.

Some things in life should stay human