Developer builds playable Paper Route game with GPT-6 Astra
Using Astra for code, Blender assets, rigging, animation and the browser build, the developer made a complete playable game in 90 commits and 39 hours, and concludes that knowing how to change the first version is what matters.
This guy didn't use GPT-6 Astra to make a game demo. He used it to build an actual game. And the surprising part? The game itself wasn't the hard part. GPT-6 Astra handled the code, Blender workflows, simulations, tests, 3D assets, animations and browser build. But getting from: “Look what AI made” to: “I actually want to play this again” required a completely different workflow. He built Paper Route, a modern Paperboy-style delivery game, through separate workstreams: → Game mechanics → Art direction → Blender asset generation → Custom 3D models → Rigging and animation → Review renders → Experiments and branches → Weather and environmental effects → UI and website polish The biggest lesson: 80% of the work wasn't building the game. It was fixing meshes, improving animations, testing mechanics, reviewing renders and adding the tiny details that give the game personality. Astra could generate the first version. Taste turned it into a game. The workflow is even more interesting: ChatGPT for concepts → Meshy for custom models → Blender/Python for assets → Astra for code, rigging and iteration → browser for constant testing. 90 commits. 39 tracked hours. 1.56B tokens. And a playable game at the end. AI can now compress the distance between idea → prototype → playable product. But the moat isn't one-shot generation. It's knowing what to fix after the first generation. The full build log breaks down exactly how he did it. Worth studying if you're building games with AI.
This post quotes another post: x.com/2007945971788021760/status/209877363124985…