Bijan Bowen gives GPT-5.6 Sol and Fable 5 the same full-stack game brief, including persistent application state, a browser interface and non-player characters powered by a local model. Both agents move from planning through implementation with limited manual intervention.
Bijan Bowen uses iterative feedback, browser testing and deployment checks to expose problems that a first code pass misses. The stronger workflow depends on agents inspecting their own output, repairing gameplay and interface issues, and validating the result on mobile as well as desktop.
Bijan Bowen finds that capable coding models can now coordinate a substantial build, but the demonstration still rewards clear specifications, observable test loops and human review at meaningful milestones.
Watch the original on YouTube