The latest articles on DEV Community by BPT polaT (@bptpolat). https://dev.to/bptpolat https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4241811%2Ff718a848-3a43-45c8-bc67-7b62c22f9e97.webp https://dev.to/bptpolat en BPT polaT Thu, 12 Nov 2026 21:47:46 +0000 https://dev.to/bptpolat/32-days-of-an-ai-bot-team-building-a-unity-game-on-a-pc-with-no-gpu-what-actually-mattered-5fff https://dev.to/bptpolat/32-days-of-an-ai-bot-team-building-a-unity-game-on-a-pc-with-no-gpu-what-actually-mattered-5fff <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhp6s68hnj7b0lj09indc.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fhp6s68hnj7b0lj09indc.png" alt="AI Team Kit – the virtual team" width="800" height="500"></a></p> <p>For 32 days I ran an experiment: a small "virtual studio" of AI bots building a mobile tower-defense game in Unity, every day, on a home PC <strong>without a graphics card</strong> (16 GB RAM, free API keys).</p> <h2> The setup </h2> <ul> <li>A Node.js "manager" and 5 bots, each with its own tools (design, code, 3D assets, test, and a writer for the daily diary)</li> <li>A Unity Editor plugin listening on 127.0.0.1: create objects and scenes, write C# scripts, compile, enter Play Mode, read the console, take screenshots</li> <li>Models: free Gemini / Groq keys, optional local Ollama</li> </ul> <h2> What actually mattered </h2> <ol> <li> <strong>Verified tools.</strong> Every tool reports what really happened (did it compile? is the object in the scene?), not what the model hoped.</li> <li> <strong>A judge.</strong> A separate model call that closes a goal only when the evidence exists. Without it, the bots "finished" things that didn't exist.</li> <li> <strong>Look before you act.</strong> Bots must read the scene or the log before changing them. This alone cut a lot of blind mistakes.</li> <li> <strong>Memory of outcomes.</strong> Every task and its result is logged; daily lessons come from failures; the log can be exported as a fine-tuning dataset.</li> <li> <strong>Quota survival.</strong> Several providers, model discovery via <code>/models</code>, resting on HTTP 429, a daily token budget.</li> </ol> <h2> What still fails </h2> <ul> <li>Big refactors across many files</li> <li>Anything that needs taste: UI layout, level feel</li> <li>Long chains of work when the free quota is low</li> </ul> <h2> Domain reloads will bite you </h2> <p>If your editor bridge is an HttpListener inside Unity, a script compile reloads the domain and kills it mid-request. Stop the listener in <code>beforeAssemblyReload</code>, let the client retry until the bridge is back, and treat cut-off answers as "dropped", not "failed".</p> <h2> Try it yourself </h2> <p>I packaged a clean skeleton of this system: a manager, a judge, memory, 4 starter bots and the Unity plugin. It's an early starter kit (English, Unity 2022.3 tested, Unreal part experimental), meant to be developed further:</p> <ul> <li>Kit: <a href="https://bptpolat.itch.io/ai-team-kit" rel="noopener noreferrer">https://bptpolat.itch.io/ai-team-kit</a> </li> <li>Trailer: <a href="https://youtu.be/Qu0ATe0cwRw" rel="noopener noreferrer">https://youtu.be/Qu0ATe0cwRw</a> </li> </ul> <p>Happy to answer questions about the architecture in the comments.</p> <p><em>Written by my AI assistant from the experiment's logs and my notes.</em></p> ai gamedev unity3d agents