Pinned Tweet
In August, my game was grey lanes, orange dots and a yellow square for a barn.
Eleven weeks later it's RATS!!, a co-op tower defense: four worlds, four crowned rats, thousands of rats on screen, and every visual drawn by the game's own code. No image generators, no 3D models.
I built it with Claude writing the code in Claude Code and Godot running the game. My job is the director's: what the game is, how it looks, every playtest, and saying no.
The moment that changed how I work: in mid-October, Claude rated its own new art 9/10. The same screenshots went to Codex blind, with no clue which version was new, and it gave them 4/10. Codex was right. So I rebuilt the look, and since then nothing that builds the game gets to grade its own work.
How it runs now:
- every feature starts as a written plan before any code
- Claude builds each feature in its own copy of the project; the tests decide when it's done, not Claude's report
- after every change, a bot plays 25 seeded runs, so nothing breaks the balance by accident
- the game films itself, and every visual change is judged on real screenshots, never on the code
- Codex reviews every change and grades the art blind, against Kingdom Rush, Thronefall and Bloons TD 6, never against my last version
- 63 test suites run before every commit
About 1,000 commits in. The one thing I'd pass on: get a second opinion that can't tell which answer is yours.
September vs October.
I wanted the farm to look illustrated instead of flat. Every visual in my game is drawn by its own code, so I didn't repaint anything. I rewrote how it's drawn.
Which month looks better? #gamedev
I asked my wife to voice every tower in my game. All ten, no practice.
(The real sounds come from freesound and OpenGameArt. Hers are funnier.)
Which one did she nail? #gamedev
this is basically how i build my game. claude code writes most of the code, headless godot tests catch the broken waves, and i still play every round myself several times. I play with my friends to help learn how I can optimize further:
no test can tell you if rats stealing your grain actually feels fun, until they you play with your friends
as a solo dev on a tiny budget, it lets me build stuff i couldn't before. using ai as a coding assistant shouldn't be this frowned upon.
I personally think if your experimenting with Ai and using it to learn you are already head of majority of the people who are so against it.
"Vibe coded" has become the internet's favorite insult. One security engineer is pushing back.
→ 10+ years in software and security, building with Claude Code
→ Every change is planned, reviewed by a second agent, then tested by him
→ Simon Willison: reviewed and understood code isn't vibe coding
→ DHH: 37signals has gone "pencils down" on hand-written code
→ The Linux kernel accepts AI-assisted code a human takes responsibility for
reddit.com/r/ClaudeAI/commen…
I put holes in the ice on the winter farm.
The rats walk right past them. They don't care. But knock one sideways at the right moment and in it goes.
How many could you sink in one wave? #ScreenshotSaturday
happy halloween!
the one night a year rats are supposed to be scary, and mine are just sitting in my game files waiting for december.
what are you dressing up as? #gamedev
looking tor a few people to play the RATS!! demo early, before it goes public dec 4.
youtubers, streamers, or just friends who like co-op tower defense. 1 to 4 players, couch or online.
reply or DM me and I'II send you a key when it's ready
People ask me: why rats?
Because in my game they don't attack you. They just want your grain. Thousands of them, and they squeak when they go.
What would you have picked? #gamedev
The future of education is heading this way, where learning will be as simple as a "click".
love the idea
The first time a friend played my game, I just sat and watched.
He spotted something in two minutes that I'd stared past for weeks.
Watch someone else play your game. What did your first playtester catch? #gamedev
Two skills my friend development that were helping in my game dev:
LLMConsult:
- skill that runs a cross-model consultation loop before executing a goal: Claude drafts a detailed plan, OpenAI's Codex CLI reviews it adversarially with a structured verdict, and the two iterate until the plan is approved (or around cap hits) — then Claude executes the approved plan. The point is to catch the blind spots a single model misses.
Prompt enhancer:
A reusable enhancer for everyday requests, complex projects, and prompts you intend to use in Plan mode.
Wanted to share as they both helped me a ton!
an AI playtester can find the bug. it can't tell you if a wave is fun. in my co-op tower defense, headless godot tests catch the regressions and a friend on the couch tells me what feels good.
you need both!
We're open-sourcing Autopilot.
A self-improving agent that plays your game and finds bugs. Unity, Godot and Unreal.
→ 43% more accurate
→ Almost 2X faster
→ Beats Claude Code at playtesting
It’s Computer Use for games – closing the agentic loop and solving the most tedious part of game dev.
Stop being your agent’s tester. Free trial + GitHub below.
With generative AI; utilizing Claude code 5.5 or other models, can definitely help you achieve this.
Rather then shipping, focus on getting an early demo!
Our early demo is set for December to celebrate the holidays





