Tweaking my game’s space sky with Claude Code took ~30 min per round: ask, wait, test, repeat.
So I had it build a Backdrop Lab: live sliders for every sky parameter in Play Mode, saved as named profiles.
Building the sky layer by layer 👇
#gamedev#indiedev#unity
I gave Claude Fable 5.2 one mission: make my capital ships feel alive.
It grilled me first. Then: heartbeat lights built into the hulls, repair drones that weld and hide under fire, and lights that die with the ship.
Before/after 👇
#gamedev#indiedev#Unity#ClaudeCode
I made my AI agent ask permission before every ElevenLabs generation. I became the bottleneck.
Now: “Use up to X credits, including retries. Stop at the limit.”
Another win: generate a short conversation, then split it into clips. Better context, fewer steps.
Boss system progress 🚀
Same hull, swap machine guns for lasers. Separate weapon and behaviour definitions make trying new combinations easy.
For this demo, 30% health means it starts circling and firing the railgun 😄
#gamedev#indiedev#unity
The coolest part of building my game with AI? Watching the tools evolve too.
My assistant built a Weapon Range to compare weapons. When I asked for a charged plasma weapon, it updated the range without being asked.
Small details. Faster iteration. 🚀
@faqdev The same way you trust your boiler when you go to shower, to not suddenly raise temp and burn you.
You create the mechanism to ensure what you build is doing it right, test it, and learn to trust it.
@mattpocockuk Could this risk addressing a symptom rather than the underlying problem?
A file-size limit seems like a useful guardrail, but how do you ensure splitting files improves the architecture rather than just satisfying the threshold?
@fadihares95@unclebobmartin Basically it creates an indexing json file, not so different than an IDE indexing, and updates using watcher by diffs.
The result is a drillable diagram.
I've recently added MCP for agents to use it for walks.