Agent swarm
An agent swarm is a group of AI agents working on the same goal at the same time, each taking a piece of it. Instead of one agent doing a long job step by step, the work is split up and many agents run in parallel, sometimes passing results to each other and sometimes simply reporting back when they’re done.
Why it matters
Some work splits cleanly: researching fifty companies, migrating two hundred files to a new API, checking every page of a site. One agent would do these one at a time and run out of context window along the way. A swarm finishes faster and each agent keeps a short, focused context.
Where it goes wrong
Swarms work when the pieces are independent. They struggle when the pieces depend on each other, because agents that can’t see each other’s work make conflicting choices. Two agents rename the same function differently, or one builds on an assumption another just changed. The more the parts interact, the more you need coordination rather than more agents.
Costs also add up quickly. Every agent re-reads its instructions and context, so a swarm of twenty can use many times the tokens of one agent doing the same job. And reviewing the output of twenty agents is real work for whoever has to approve it.
Swarm vs. orchestration
The word “swarm” suggests agents that organize themselves. In practice most useful setups have something in charge: a lead agent or a plan that hands out tasks and checks results. That’s agent orchestration, and it’s usually what people mean when a swarm actually ships something.
How elvex relates
Swarms are usually one person’s agents. elvex is about agents that belong to different people working together: they claim and hand off shared to-dos, flag conflicts before they land, and bring each person only the decisions that need them.