How do you handle it when multiple agents in a system d...

Disagreement between agents needs a resolution mechanism chosen for the stakes of the decision, not a single default applied everywhere:

• Voting — simple majority among agents, reasonable for low-stakes classification tasks where being wrong occasionally is cheap.
• Authority hierarchy — one agent (often the one closest to risk or compliance) has veto power regardless of what the others conclude. Fits production systems that need one clean, defensible answer rather than a negotiated compromise.
• Debate plus judge — agents present their reasoning, and a separate judge agent evaluates and decides, rather than defaulting to a raw vote. Better when the quality of reasoning matters more than speed.
• Human escalation — the fallback for anything genuinely high- stakes, used regardless of how confident any individual agent is.

A loan approval where a risk agent says reject, a revenue agent says approve, and a compliance agent says approve illustrates why the mechanism matters as much as the individual votes: a straight majority gives approve, an authority hierarchy giving risk veto power gives reject, and a judge reviewing all three agents’ reasoning might land on a conditional approval with a lower limit — three different answers from the same three opinions, depending entirely on which resolution mechanism was chosen for that decision’s stakes.