When people hear "AI-generated quests" they picture a single model that takes your goal and spits back a to-do list. We tried that first. It produced confident, well-written, completely generic advice — the kind you could get from any article. Useless.
One model, too many jobs
The core issue is that a good quest requires several different kinds of reasoning at once: understanding your psychology, knowing your current capability, calibrating difficulty, keeping continuity with what you did yesterday, and writing it in a voice that lands. Asking one prompt to do all of that means it does all of it adequately and none of it well.
Splitting the work
So we split the engine into five specialised roles — what we internally call the 5-Brain System. Brain 1 owns the Daily Board — your LOOP, INVEST, and GYM missions for today. Brain 2 is the Chaos Side Quest Engine, generating bonus-board candidates. Brain 3 evaluates custom quest submissions and writes your weekly reports. Brain 4 is the Recovery Engine — the mission you get after a Hemorrhage, built to interrupt the pattern, not lecture you about it. Brain 5 generates your Gym Mode training programme around your declared equipment and class.
Each brain has a narrow job, a focused prompt, and a clear contract for what it passes to the next. Nothing crashes because one brain tries to also be another brain.
Why specialisation wins
There is a reliability benefit too. When one stage misbehaves, we can see exactly which one and fix it in isolation, instead of tuning one giant prompt and praying nothing else regresses. Narrow components are debuggable. Monoliths are not.
It is more moving parts and more orchestration to maintain — but the quality difference is the entire product. A habit platform whose quests feel generic has no reason to exist. The 5-Brain System is how we make sure they never do.