# What is Orchestration Engine?

Also called workflow engine, orchestrator.

An orchestration engine is infrastructure that coordinates multi step processes, deciding what runs next, tracking each step's state, and handling retries, timeouts, and recovery after a crash. It separates the definition of a process from the machinery that reliably executes it. Common examples include workflow engines for business processes, job schedulers for data pipelines, and coordinators for multi step agent runs.

Without an engine, multi step logic tends to be encoded implicitly across queues, database flags, and scheduled scripts. That arrangement works until something fails halfway through, at which point nobody can say which steps completed and whether resuming is safe. An engine makes the process explicit by persisting step level state, so a run can be inspected, resumed, or compensated deterministically.

Engines vary along several axes. Some define processes as code and others as declarative graphs. Some guarantee durable execution that survives process restarts, while others simply retry from the beginning. Some are centralized services with their own storage, and others are libraries running inside the application. The right choice depends on how long processes run and how costly a repeated step is.

A distinction worth keeping clear is orchestration versus choreography. Orchestration means a coordinator explicitly directs each step. Choreography means services react to events published by others with no central director. Orchestration gives visibility and straightforward error handling; choreography gives looser coupling. Most real systems mix the two, and which to prefer is an ongoing debate in distributed systems design.

For agent systems the term is overloaded and worth disambiguating. Some use it for the infrastructure layer that keeps long agent runs alive across restarts and retries. Others use it for the reasoning layer where a planner agent delegates to sub agents. The two solve different problems, and a system commonly has both: a durable engine underneath and a model driven control flow above it.

## Key points

- Coordinates multi step processes and persists step level state.
- Handles retries, timeouts, and recovery after a crash.
- Separates process definition from reliable execution machinery.
- Orchestration directs steps; choreography reacts to events.
- In agent contexts the term covers both infrastructure and planning.

## In practice

A customer onboarding process involves creating a workspace, importing records, running an agent to summarize them, and sending a welcome message. Defined in an orchestration engine, each step records its outcome. When the import step fails partway through, the engine retries only that step with backoff, leaves the completed workspace creation untouched, and after three failures marks the run for review rather than sending a welcome message about data that never arrived.

## Related terms

- [Workflow](/en/glossary/workflow)
- [Durable Execution](/en/glossary/durable-execution)
- [Queue](/en/glossary/queue)
- [Scheduled Job](/en/glossary/scheduled-job)
- [Retry policy](/en/glossary/retry-policy)

[Back to the AI Glossary](/en/glossary)