Council vs Debate

Every AI app gives you one model's answer. BEAM gives you the seam between several. Claude, GPT, and Gemini are trained differently, weighted differently, and wrong in different places — and the disagreement between them is information no single model can give you. The two multi-model modes are two ways of extracting it.

Council: parallel, independent

>council should the talk open with the demo or the story?

Each model answers the same question without seeing the others. You read three answers side by side — ◆ Claude, ◇ GPT, ○ Gemini. Where they converge, you have something solid. Where they split, you've found the actual decision, and it's yours to make.

Council is the default multi-model move: cheap to read, fast to run, and honest about disagreement because nothing is smoothed over by a synthesis step. When you do make the call, council align <model> records which answer you sided with — over time BEAM gets a read on which models earn your trust, and shows you, while still showing every position in full. The minority view has been right before.

Debate: adversarial, sequential

>debate is a live demo worth the risk in a 30-minute slot?

Debate makes the models engage each other: initial positions, two rounds of critique, final positions, then one synthesis. It's slower and costs more than council, and that's the point — positions that survive being attacked are worth more than positions that were merely stated.

Debate also has formats for specific shapes of question: blind (independent answers merged without anchoring), lenses (each model argues from a different worldview), redteam (one proposes, one attacks, one judges — the stress test), stakeholders (models argue as the people affected), godin (start from the smallest audience that matters). debate help lists them.

When not to multi-model

ask is right when the answer is routine, when speed matters more than depth, or when you need a lookup rather than a judgment. Multi-model modes are for decisions — the moments where being wrong is expensive and disagreement is worth paying to see. A day of BEAM use is mostly ask with a handful of council moments; a big decision might earn one debate.

And when you make the call, record it: decide <choice> --why <rationale> captures your decision against the model positions that informed it, so six months from now you know not just what you chose but what the dissent was.

See also: Reference: council, debate, ask, decide · Concepts: Flagship tier