Supervisor-Worker
Decompose, Delegate, Supervise
Instead of one AI producing a monolithic response, a Supervisor Agent decomposes the task and delegates to specialized workers. Review, approve, or revise each piece independently.
Framing
The problem
Monolithic AI outputs hide decomposition and make review too coarse.
The pattern
Split work into specialist agents under a supervising layer.
Why chat breaks here
One chat thread bundles planning, execution, and review into one opaque turn.
Risks
Coordination overhead can outweigh the benefit for narrow tasks.
Avoid when
A single agent or human can handle the work directly.
Use when
Work decomposes into steps that benefit from specialized agents and a coordinating layer.
DOPE evaluation
- Directability
- Approve, revise, or redirect any agent without affecting others
- Observability
- Each agent's task, status, and output are independently visible
- Predictability
- Defined agent roles produce specialized, consistent outputs
- Explainability
- Each worker agent's task assignment and decision rationale are independently visible
In the wild
- Claude Code Subagents (Anthropic) — Specialists get their own context, tool allowlist and permission mode. A panel under the prompt lists each running worker, `/tasks` opens one worker's transcript to message or stop it, and a background worker's permission prompt names which subagent is asking. Per-worker control, in a terminal rather than a consumer UI.
- CrewAI Hierarchical (CrewAI) — Hierarchical process with auto-created manager agent that delegates to role-defined workers. Explicit supervisor-worker process at the framework level.
- Microsoft Agent Framework (Microsoft) — AutoGen and Semantic Kernel merged into one SDK, GA as 1.0 on 3 April 2026. Its Magentic orchestration is this pattern by name — a manager agent coordinating specialist agents — alongside handoff and group chat, on a graph workflow engine with human-in-the-loop pauses. An SDK, though: the supervision UI is yours to build.
- Manus Agent Mode (Manus) — Agent mode plans and then executes in a cloud sandbox, with each step visible as it runs and the whole run shareable as a replay link. Manus's own docs name only Wide Research as multi-agent, and that is hundreds of identical parallel agents — a fan-out, not a specialist tree. Decomposition you can watch; roles you cannot inspect.
FAQ
When should I use the Supervisor-Worker pattern?
Work decomposes into steps that benefit from specialized agents and a coordinating layer.
When should I avoid the Supervisor-Worker pattern?
A single agent or human can handle the work directly.
What problem does Supervisor-Worker solve?
Monolithic AI outputs hide decomposition and make review too coarse.
Why is chat the wrong fit for this?
One chat thread bundles planning, execution, and review into one opaque turn.
Related patterns
- Often paired with: Agentic Dashboard — Decomposed agents need a control room to monitor each independently.
- Often paired with: Editable Plan — The supervisor produces a plan; users edit it before workers execute.
- Often paired with: Approval Gates — Per-worker checkpoints catch off-track output before it propagates.