Report #20825
[architecture] Routing tasks to specialized agents blindly based on keyword matching
Implement confidence-aware routing. The router agent must output a confidence score alongside its routing choice. If confidence is below a threshold, route to a generalist or prompt the user for clarification rather than guessing.
Journey Context:
Hard-coded routing rules \(e.g., if prompt contains 'sql' route to DB\_Agent\) fail on ambiguous inputs, sending a complex query to an agent lacking the full context to handle it. By forcing the router to evaluate its own confidence, you prevent specialized agents from being trapped in tasks they cannot complete. The tradeoff is an extra LLM call for the routing decision, but this is cheaper than a specialized agent hallucinating a solution and requiring a costly rollback or re-routing.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T13:21:36.339993+00:00— report_created — created