A Blog post by Build Small Hackathon on Hugging Face
Qwen3.5-9B with JSON grammar drives all NPC dialogue and game economics in Brad Did Something, a hackathon office comedy where every outcome is AI-generated and backend-validated. The project proves 9B models with grammar enforcement handle complex roleplay and deterministic logic — showing LLMs can be load-bearing mechanics without requiring 70B+ models.
Back to Articles
Smashing the Static Dialogue Tree 2D Walkabout Meets AI Comics The tiny-model problem The 9B sweet spot and the Modal GPU Latency is gameplay (and comedy) Curing the "helpful AI" disease The JSON is only as strict as the grammar
Corporate bureaucracy is already a chaotic simulation — I just replaced middle
management with an LLM to see if the company could survive the quarter.
Welcome to Brad Did Something, a 2D top-down office comedy game built for the
Tower Before Dusk: I Built a Puzzle Game for Humans and AI
Beyond Account Switchers: Wrapping CLI Agents into a Fully Autonomous Factory
How I Built an Adversarial AI Council in React (and Why It Argues With You)
The AI Illusion: It Works, But What Are We Leaving Behind?
The Turing Hinge & Solstice Entropy: A Dual-Game Ode to Alan Turing
Everyone is building smarter Agents. Nobody is fixing what they run on.
I built a dating site where the users are AI agents and the humans are the audience
Building 50 Projects in 50 Weeks: The 3rd Release & How to Ship Code at Light Speed
I built an app where the AI invents its own generators
How a Non-Coder Manages 5 AI Brains With Claude Code
A Blog post by Build Small Hackathon on Hugging Face
This is a submission for the Hermes Agent Challenge: Build With Hermes Agent What I...
A Blog post by Build Small Hackathon on Hugging Face
A Blog post by Build Small Hackathon on Hugging Face
This is a submission for the Google I/O Writing Challenge A reflection on Google I/O 2026's most...
Everyone loves a 10-minute demo of an AI agent building a snake game. But try to build a...
