In April 2023, I tried to play Fantasy Age with ChatGPT. I gave it the rules. It couldn’t follow them. I reminded it of the rules. It still got them wrong. Not because it was stupid — because it couldn’t hold persistent state across a session.

That frustration became a two-year project.

The Realization

The realization was simple: AI systems can’t reliably follow rules. Not because the models aren’t smart enough, but because there’s no infrastructure to hold the rules in a way the AI can’t forget, ignore, or hallucinate around.

What I Built

So I built that infrastructure.

Truth Nodes hold project rules that persist across every session — the AI reads them, but can’t modify them. The Context Gate forces the AI to read them before it’s allowed to edit anything. The Constraint Engine checks every action against those rules before execution. The Memory System preserves decisions across sessions so the next agent inherits what the last one learned.

It’s the system I wished I had for D&D: an immutable rulebook that the AI has to consult before every action, with a referee that blocks illegal moves. Turns out, that’s exactly what software development needs too.

The Outcome

The tabletop RPG never worked out. But the governance framework now manages 9 projects across 7 programming languages. Sometimes the best ideas come from the most frustrating failures.