# AI Agent Routing AI Agent Routing is the mechanism that maps user intent to the appropriate [[AI Agents|agent]], [[AI Agent Panels|panel]], or team. It determines which agent handles a given request, ensuring that the right expertise is applied to the right problem. ## Two-Tier Architecture Effective routing uses two tiers to balance speed and accuracy: **Tier 1: Lightweight hints (always loaded)** A routing table embedded in the bootstrap configuration (e.g., `AGENTS.md`) that maps common domains to agents. Covers obvious cases without loading additional context. Costs almost zero tokens. **Tier 2: Dynamic receptionist (on-demand)** For ambiguous requests, a full [[Receptionist AI Design Pattern|Receptionist]] skill loads, reads the agent registry, classifies intent, and routes dynamically. More expensive but more accurate. ## Classification Order 1. **Direct name match**: User mentions an agent by name 2. **Domain match**: Request clearly falls in one agent's expertise 3. **Panel match**: User asks for multi-angle feedback 4. **Team match**: Goal requires multiple agents collaborating 5. **Ambiguous**: Could go to 2+ agents; one clarifying question, then routes 6. **No match**: Handles generically or suggests creating a new agent ## Design Principles - **Never cache the registry**: Always read fresh. Agents get added, removed, or modified. Stale routing produces wrong results. - **Prefer specificity**: Route to the most specific agent possible. A business pricing question goes to the Pricing Advisor, not the generic Strategist. - **Ask before guessing**: When ambiguous, one clarifying question is cheaper than routing to the wrong agent and wasting a full exchange. - **Routing is not orchestration**: Routing picks the destination. [[AI Agent Orchestration]] manages the collaboration once agents are active. ## Connection to Context Engineering Routing is the first major [[Context Engineering]] decision in any multi-agent session. The choice of which agent to activate determines which context gets loaded. Good routing keeps context lean and relevant. Bad routing wastes the token budget on irrelevant context. ## References - ## Related - [[AI Agents]] - [[AI Agent Identity]] - [[AI Agent Panels]] - [[AI Agent Orchestration]] - [[AI Assistant Architecture]] - [[Receptionist AI Design Pattern]] - [[Context Engineering]] - [[Prompt Lazy Loading AI Design Pattern (PLL)]]