Fleet of Agents

An operational label, still without a universal taxonomy, for a set of autonomous or semi-autonomous agents provisioned as a worker pool. Each agent may receive its own task, context, workspace, and budget while a control plane distributes work, limits concurrency, tracks state, and collects results. The benefit comes from decomposing independent tasks and executing them in parallel, provided that coordination costs remain lower than the distributed work.

Fleet, swarm, and multi-agent system describe different aspects. Fleet emphasizes operational scale and governance; swarm emphasizes decentralized coordination through local rules and emergent behavior; multi-agent system covers architectures in which multiple agents coordinate a solution. A fleet may use a central orchestrator and workers that never exchange messages with one another. Production requires per-agent identity, isolation, quotas, idempotency, retries, prompt and tool versioning, traces, evals, and stopping conditions. Without these controls, parallelism merely amplifies duplicate work, failures, and cost.