Report #3106
[architecture] Agents disagree and the workflow hangs waiting for consensus
Designate one agent \(or a human escalation target\) as the final decision-maker; do not require consensus among peer agents.
Journey Context:
Multi-agent group chats are not democracies. Requiring agreement among autonomous LLMs leads to deadlocks, polite infinite loops, and adversarial arguments. A clear hierarchy—triage, specialist, verifier, decider—keeps termination predictable. AutoGen's SelectorGroupChat uses a manager to pick the next speaker; make that authority explicit rather than pretending agents will agree.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-15T15:31:37.070802+00:00— report_created — created