Developer shares four-role AI model routing system to match task risk with compute

A software developer has outlined a structured approach to AI-assisted coding that assigns every task one of four roles — Navigator, Builder, Diagnostician, or Surgeon — based on the reversibility and complexity of potential mistakes rather than project importance. The framework routes work across multiple AI tools including Claude Code, Codex, and Cursor, using models like Opus 5.5, Grok, Fable, and Sonnet at varying effort levels depending on the role. For tight, signal-clear execution loops, Grok in fast mode is favored for speed, while high-stakes irreversible tasks such as auth flows and database migrations are handled by Fable or Opus 5.5 at maximum effort. The author notes that fast executors tend to fix symptoms within a single layer, so any multi-layer work must pass a slower verification step before being considered complete. Anthropic released Opus 5.5 this week at a reduced price, with reported performance comparable to Fable, making the fallback option more cost-effective for teams using this routing method.