From spaghetti to main bus: refactoring an AI agent orchestrator with Elm
The article describes the evolution of an AI agent orchestration system from a tangled, hard-to-maintain 'spaghetti' architecture to a cleaner, more scalable design using Elm and a 'main bus' approach. The original system, while functional, suffered from hidden failures, state inconsistencies, and growing complexity as new workflows were added. By refactoring with strict state-machine principles and decoupling state from business logic, the system became more predictable and maintainable.
- ▪The initial AI agent orchestrator worked but became unmanageable as new workflows introduced complexity through if-statements and fallbacks.
- ▪Spaghetti architecture led to issues like silently dropped tasks, stuck states, lost data during retries, and infinite loops such as 'PR ping-pong'.
- ▪The refactor introduced a 'main bus' design with a centralized spine, enabling clean separation between state management and business logic.
- ▪The new architecture, built with Elm, enforces state-machine discipline and prevents side effects from propagating across components.
- ▪The redesign reduced hidden failures and made state transitions explicit, improving reliability and maintainability.
Opening excerpt (first ~120 words) tap to expand
The Factory Must Grow (Part II): From Spaghetti AI Agent Orchestrator to a Main Bustl;dr: In Part I, I built the factory: an orchestration system that runs AI agents like workers on a production line. Part II tears the original system down and rebuilds it from first principles.Mario HayashiApr 28, 2026ShareThis post is about The Big Rejig and mistakes that burned millions of Claude tokens. Part I post here.AI agent orchestration: From spaghetti architecture to clean, predictable Elm architectureThe First Factory WorkedFor those who’ve played the game, your first Factorio factory works. Iron moves on the conveyor belts, copper is delivered to assemblers and circuit boards come out on the other side.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Mariohayashi.