Definition
Agent orchestration
Agent orchestration is the coordination of multiple AI agents — and the hand-offs between them — to complete work no single agent owns end to end. A planner decomposes a goal, specialist agents execute sub-tasks, and results are routed and merged. The hard part is not running the agents but governing what each is allowed to do as they hand work back and forth.
Key takeaways
- Agent orchestration coordinates multiple agents and their hand-offs: decompose a goal, route sub-tasks, merge results.
- Common shapes are a supervisor delegating to workers, peer agents negotiating, or a fixed multi-stage pipeline.
- Per-agent quotas under-govern multi-agent work; the real risk is each consequential action, wherever in the chain it originates.
- Planoda governs the action, not just the agent — every step in an orchestrated flow routes through propose/approve and one audit trail.
As soon as you have more than one agent, you need a layer that decides who does what, in what order, and what happens when one agent's output becomes another's input. Orchestration covers task decomposition (splitting a goal into sub-tasks), routing (sending each sub-task to the right specialist), hand-offs (passing context and partial results between agents), and aggregation (merging the pieces back into one coherent result). Patterns range from a single supervisor delegating to workers, to peer agents negotiating, to a fixed pipeline.
The governance question gets sharper, not softer, as agents multiply. Coarse per-agent quotas (how many runs, how much spend) are not enough when one agent can trigger another: the real exposure is the individual consequential action — the status change, the bulk archive, the external API call — wherever it originates in the chain. Notion's External Agents API, for instance, lets third-party agents act inside a workspace, which makes per-action control the thing that actually bounds risk.
Planoda's answer is to govern the action, not just the agent: every consequential action in an orchestrated flow — whichever agent or hand-off produced it — routes through propose/approve and lands in one audit trail, so multi-agent work stays attributable instead of becoming an opaque relay of automated changes.
Related terms
- AI AgentAn AI agent is a software system that uses a large language model to pursue a goal across multiple steps — reading context, choosing tools, and taking actions — rather than answering a single prompt. In a work platform, agents triage issues, draft updates, and execute multi-step tasks as autonomous teammates, bounded by the permissions and approvals their operators set.
- Agentic WorkflowAn agentic workflow is a process in which one or more AI agents carry out a multi-step task with some autonomy — planning, calling tools, and acting on results in a loop — rather than a human driving each step. The agent decides the next action toward a goal, within boundaries its operator sets, turning AI from a single-response assistant into a worker.
- Agent governanceAgent governance is the set of controls that make an AI agent's actions safe, attributable, and reviewable: human approval gates on consequential actions, an immutable audit trail of who approved what, role-based capability limits, and spend controls. It is the difference between an agent that suggests and one you can trust to act.
- Propose / Approve (AI Governance)Propose/approve is a governance pattern for autonomous software: instead of executing a consequential action directly, an AI agent emits it as a proposal that a human or policy must approve before it runs. It keeps fast, read-only work autonomous while gating destructive or irreversible operations — the practical way to give agents real power without surrendering control.
- Assignable agentAn assignable agent is an AI agent modeled as a first-class workspace member: it has an identity you can @mention, assign issues to, and schedule, just like a human teammate. Instead of living in a separate chat panel, it appears in the assignee picker and the activity feed — so delegating to AI uses the same gestures as delegating to a person.