💻 Programmer/crafter 🇹🇭 Emigrated to Thailand 🎮 Making a game
Wishlist on Steam ➡️
- Tweets 1,890
- Following 218
- Followers 245
- Likes 3,093
In the process of recycling my manual, non-AI work from Ardaria into that new rust envelope, I'm porting main shaders I did for the vegetation, it took me months to get it right, one hour to port to Three.js
4
24
I spent 15 years loving the act of writing code. Ardaria is especially personal: I built its systems line by line in Unreal. Letting AI help build it feels wrong, and a little sad. But I don’t want to leave that world behind, so I’m testing a lighter Rust engine
2
5
83
I'm waiting for dev day, but after worked with Opus 5.5 for a day (after 4 months of codex gpt 5.3 to 6), OpenAI feels left behind.
The usage is insanely better, the 'straight to the point' of opus PLUS the TPS, for WAY LESS usage ???
2
43
TO ALL DIE HARDS of formal proof, this is actually useful for agents. Opus 5.5 fianlly solved a problem I had for weeks regarding voxel terrain generation for a tech test "i" am doing in rust where I'm vibing the great principles of the algo, don't ask me to read more code
32
Fun fact I was working on writing TLA+ algos ahead of the coding to get agents focused on the logic.
I'm using Quint because it's forces you to write state machines, which are easy to reason about.
I wasn't sure to be on the right track, I feel a bit reassured now !
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached.
TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt.
I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted.
Is formal verification the future of coding (or at least, bug finding)?
1
50
Running one gpt5.6 Sol agent continuously was around 1%/h
The same gpt6-sol agent now runs at 0.5% / h
Metrics with 5/6 agents active, including some Astra. Usage seems way better on Codex now.
Should I still drop my second codex x20 for one Claude x20?
1
1
127
We all gonna die, so let's make cool stuff
ゲームエンジンを作り始めた…
車輪の再発明？
無理やり肯定すると
→AIでみんながゲームを作り出した
→競争力↑
コミュニティを持つ人達は一定の成功はしそう
自分のゲームに最適なエンジンを作ればこれまでにない体験ができるはず！
※結局AIに駆逐されそうだが…
#ゲーム開発 #indiegame
1
3
84
ゲームエンジンを作り始めた…
車輪の再発明？
無理やり肯定すると
→AIでみんながゲームを作り出した
→競争力↑
コミュニティを持つ人達は一定の成功はしそう
自分のゲームに最適なエンジンを作ればこれまでにない体験ができるはず！
※結局AIに駆逐されそうだが…
#ゲーム開発 #indiegame
7
3
64
3,680
Created an UX to iterate on Specs, even Astra doesn't get the software specifications right on the first try, so I made different feedback categories with one click + a voice feedback if necessary.
When the spec overscopes or is badly said, I just press a button.
1
60
My pretty formatting when codex writes. I feel formatting is critical to quickly undertand and decide
1
2
49
I can't stand anymore micro-managing individual agents for code implementation. I've been trying to take my 40,000 hours of software engineering in a replayable procedure for the last 10 months, mitigated success.
But I might be able to finally solve autonomous coding?
1
1
54
Let's connect if you are working on proof for software.
I'm having some experiment with Quint, curious if you know Bend
1
18
