Agents
Multi-Agent System
Instead of one all-rounder, several specialists: one researches, one writes, one checks, coordinated through defined handoffs. This can raise quality, because each agent has a clear role, and it can multiply chaos if nobody holds responsibility. The rule from organizational theory applies here, too: first the process, then the staffing.
An example
Instead of one agent for the complaint, there are four: a researcher who gathers facts, a calculator, a drafter, and a checker who holds the result against the policy. The advantage is real: the checker finds errors the drafter doesn't see, because the drafter made them. The price is just as real: four times as many calls, four times as many places where something can get stuck, and a log you first have to learn to read.
A closer look
Several specialized agents share a task, often with a coordinating agent above them. The advantage lies in clearer responsibilities and shorter contexts per agent, the price in communication overhead and harder traceability.
Why it matters
The sober rule: multiple agents pay off where different roles lead to different results, above all in checking. A second pass with the assignment of refuting its own result demonstrably finds errors. They don't pay off as a replica of an org chart. A system in which one agent assigns tasks to another, which reports to a third, has the coordination costs of an organization and not its judgment. Start with one and justify each additional one.
Commonly misunderstood
A team of agents rarely solves a problem a single agent fails at. Usually it spreads the same error across more places and makes it harder to find.