Multi-agent / swarm

Several agents (AIs that work towards a goal on their own, step after step) on one problem at the same time. "Swarm" is the loose version: many of them exploring at once, barely coordinated — five agents each drafting a different version of the same page, then something picking the best one.

Three firms quoting the same job independently, and you take the best quote. You didn't get a better quote by thinking harder. You got it by asking three times.

Why it matters: several attempts plus a choice often beats one careful attempt, which is a genuinely new way to buy quality — with volume rather than with effort. The catch is that coordinating them costs something too. More agents is not automatically better, and past a point it's noise and a bigger bill.

Related: Agent · Parallel · Fan-out