Definition
Plain language
A setup where several AI models work together and one combines their answers.
As stated in the literature
A multi-agent inference architecture that aggregates outputs from multiple models, typically through a fixed aggregator model; contrasted with adaptive-aggregator orchestration systems like Fugu that vary which model synthesizes per task.
Why it matters: It lets you combine the strengths of several models into better answers than any single one would give on its own.
For example, three different models each draft an answer to a question, and a fourth model reads all three and merges them into one final response.
Heard on the show
“The prior work in this space — Mixture-of-Agents, GPTSwarm — uses one fixed model as the aggregator, which bottlenecks the whole system on tasks outside that model's strength.”Episode 166 — A Router That Beats the Frontier Models It Calls