Multi-agent system
Several AI agents with distinct roles, such as planner, worker and reviewer, coordinating on one task or goal.
A multi-agent system splits work across specialised agents — a planner, researchers, a coder, a reviewer — typically coordinated by an orchestrator. It helps when subtasks need different tools or context, or when parallel exploration is valuable. Costs are token volume (each sub-agent has its own context) and coordination failures. Stack Overflow's 2025 survey found 28.1% of developers building agents use multi-agent orchestration.