site banner

Tinker Tuesday for April 28th, 2026

This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.

Post your project, your progress from last week, and what you hope to accomplish this week.

If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service.

2
Jump in the discussion.

No email address required.

...have you tried asking AI?

I've been thinking for the last year or two that I needed to start engaging with AI more, but wasn't really sure how to get stuck in. I was attempting to google a blocking issue on one of my unity projects, and accidentally clicked on the AI summary result at the top... and the AI summary was extremely helpful, and got me past the problem in short order. Since then, I've been turning to it faster and faster when hitting weird blocker issues, and it has one- or two-shotted them every time so far, usually just by providing context to highlight the issue I'm blind to.

Yes, I've been making extensive use of GPT 5.4 through Github Copilot. It's been working quite well for most things.

  • It helped me a lot of getting a handle on various Unreal systems, including procedural geometry.
  • I told it to create a mechanism for measuring game performance and shutting it down if CPU and/or memory use get out of hand, to prevent freezes. It did, and I had no complaints about it.
  • I told it to write a script to automatically start the game after each prompt to validate that it still runs. It did, and even uses it semi-reliably, and when it doesn't then a quick reminder prompt is enough to get it going. Also refined that approach with copilot-readable logging. The latest addition was taking screenshots at regular intervals for the copilot to inspect, which almost works: It requires the manual step of me dragging them into the chat window.

And, latest item

  • It actually did fix the atmosphere rendering! Took several days and dozens of different angles to finally get at it, but it works now.

So yeah, AI is a pretty powerful tool for this. It just takes so much refactoring to keep the codebase in a good state when the AI has zero qualms about writing a million lines for each little fix I request, and a lot of attention to keep it from breaking things on the way.

Next up, why did my background stars particle system stop working? I'll probably bang my head against that one for a while, too.

AI is still at the state where it will sometimes confidently tell you something completely wrong, and yet for simple blocker issues this is still immensely helpful, because you can just try out what it told you, and 90% of the time you get to declare victory, and the other 10% you're just back where you started.

In fact, there's a bit of grey area in between - twice recently I've seen an AI come up with a solution which was definitely wrong or incomplete in some way, but which was much easier to fix than solving the problem from scratch would have been.

For complex issues and design issues, AI can easily paint you into a corner by generating reams of redundant/spaghetti/inflexible code that solves your immediate problem but is unmaintainable in the long run, but in general it's getting better so fast that I'm not sure how long this warning will be necessary.