Agent Orchestration
Also known as: Agent Coordination, Agent Management
Definition
The coordination of multiple agents or agent components to accomplish complex tasks. Orchestration determines which agent handles which subtask, manages communication between agents, handles failures and retries, and ensures the overall goal is achieved. Orchestration can be centralized (a supervisor or router controls agents) or decentralized (agents self-coordinate).
What this is NOT
- Not the same as a workflow engine (orchestration involves dynamic decisions)
- Not just running agents in parallel (orchestration implies coordination)
- Not the agents themselves (orchestration is the coordination layer)
Alternative Interpretations
Different communities use this term differently:
llm-practitioners
The patterns and infrastructure for managing multiple LLM-based agents, including routing requests, managing agent lifecycles, passing context, and aggregating results.
Sources: LangGraph documentation, AutoGen orchestration patterns, OpenAI Swarm (experimental)
distributed-systems
Coordination of distributed services or processes, typically involving service discovery, load balancing, and workflow management.
LLM agent orchestration borrows concepts from service orchestration but adds reasoning about when to delegate vs. execute directly.
Sources: Kubernetes documentation, Service mesh patterns
Examples
- A router that sends coding questions to a code agent and research questions to a search agent
- A supervisor that breaks a task into subtasks, assigns them to specialists, and synthesizes results
- LangGraph managing state transitions between multiple agent nodes
- A pipeline that dynamically decides whether to retrieve, reason, or respond
Counterexamples
Things that might seem like Agent Orchestration but are not:
- A single agent operating in isolation
- Hard-coded sequential steps with no branching
- Running the same prompt through multiple models and taking the first response
Relations
- overlapsWith multi-agent-system (MAS requires orchestration for coordination)
- requires agent (Orchestration coordinates agents)
- overlapsWith workflow (Orchestration may include workflow-like sequential steps)
- overlapsWith supervisor-agent (Supervisors are one orchestration pattern)
Implementations
Tools and frameworks that implement this concept:
- Amazon Bedrock secondary
- AutoGen primary
- Claude Cowork secondary
- CrewAI primary
- Dify primary
- Flowise primary
- Google Stitch secondary
- LangChain primary
- LangGraph primary
- LangSmith secondary
- Letta (MemGPT) secondary
- LlamaIndex secondary
- n8n secondary
- Semantic Kernel primary
- Superpowers (obra/superpowers) primary