Report #69582
[frontier] Agent routing loops or selects wrong agent when using LLM for orchestration
Implement a code-driven state machine \(e.g., LangGraph\) for orchestration, restricting LLMs to node-level execution only.
Journey Context:
LLM-driven routing \(where Agent A decides to call Agent B\) introduces non-determinism, high latency, and infinite loops. Deterministic graphs ensure predictable state transitions, allowing explicit retries and fallbacks at the graph level. The LLM should only reason within a node, not control the graph edges. Code-driven orchestration trades the flexibility of agent-to-agent delegation for the reliability required in production systems.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:16:40.557236+00:00— report_created — created