Skip to content

Orchestrate a multi-agent review

One answer from one model is a guess with confidence. When the cost of being wrong is higher than the cost of a few extra calls, run several and converge. That is what the AI Mastermind is for.

Open AI Mastermind (/room). Each header chip is a participant; open ⚙ Models to choose the concrete model for each seat and whether it may think before answering. Trim the roster to two or three - the default adds every reachable agent.

Pick MoA. Every participant proposes; each further layer (1-3) reads all of the previous layer’s answers - numbered and anonymous on purpose, so a smaller model judges content, not brands; then the aggregator (Σ) synthesises one reply. The header line says what the run will cost in calls before it starts - read it, then commit. ASK and DISCUSS are different: they return several answers, not one.

Give it something where perspectives genuinely differ - a design trade-off, a review of a risky change, an ambiguous spec. A question with one obvious answer wastes the mechanism.

The room returns a single answer, not a transcript to reconcile by hand. If you want the individual answers too, they are there - but the point is the convergence.

For a single request that must simply stay answered when a provider is down, you do not need orchestration - the control plane’s failover already handles that, invisibly and for free.