Agent fleet
Also known as: agent fleet management, fleet of agents, agent fleet orchestration
When you have one agent, the hard problems are prompting and evaluation. When you have ten, twenty, or fifty, you have a fleet, and the hard problems shift toward coordination: how tasks get assigned, how agents hand off to each other, how you know which ones are stuck, and how you control the total cost before it becomes a boardroom conversation.
The term has moved from technical documentation into everyday builder vocabulary in 2026 as multi-agent systems went from experimental to normal. A fleet might be a set of parallel coding agents each working a separate git worktree (branch of a codebase), a set of data enrichment agents each pulling from different sources, or a mix of specialized agents with a conductor routing tasks among them. What they share is that no single agent handles the whole job, and the team has to manage the group as a unit.
Fleet management as a practice includes centralized visibility (which agents are active, what each is doing), task queuing and assignment, cost tracking per agent, and observability tooling that spans the group rather than logging one agent at a time. Tools like Orca (agent fleet orchestrator) and AgentCenter address this category directly, and cloud providers have started building fleet management primitives into their agent services.