Orchestration patterns

12 patterns for coordinating work across processes, services, agents and people. Each pattern page lists when to use it, when to avoid it, tradeoffs, failure modes and implementations.

Reliability

Keep work correct and progressing when processes crash, calls fail or transactions span services.

Pattern	Definition	Maturity
Durable Execution	Durable execution records every step of a workflow in a persistent log so that the workflow survives process crashes, restarts and deploys, and resumes from the last completed step as if nothing happened.	Established
Saga / Compensation	A saga splits a business transaction that spans several services into a sequence of local transactions, each paired with a compensating action that semantically undoes it if a later step fails.	Established
Retry & Fallback	Retry & fallback re-attempts a failed step under a declared policy of attempts and backoff, and switches to an alternative path when retries are exhausted or the error is not retryable.	Established
Checkpointing	Checkpointing saves a snapshot of execution state at defined points, so that after a failure, pause or inspection the run can be restored from the latest snapshot rather than restarted from the beginning.	Established

Control flow

Decide what runs next, in what order and in parallel.

Pattern	Definition	Maturity
Router	A router inspects each incoming request or intermediate state and sends it to exactly one of several downstream paths, based on rules, a classifier or a model decision.	Established
Fan-out / Fan-in	Fan-out / fan-in splits work into independent parts that run in parallel, then waits at a join point to collect and combine their results before the workflow continues.	Established
State Machine	A state machine models a process as a finite set of named states and the allowed transitions between them, so that at any moment the process is in exactly one known state and only valid moves are possible.	Established
DAG	A DAG orchestrates tasks as a directed acyclic graph of dependencies: each task runs once its upstream tasks have succeeded, and independent branches can run in parallel.	Established

Coordination

Divide responsibility between services, agents and people.

Pattern	Definition	Maturity
Supervisor	A supervisor is a central coordinator that breaks a task into sub-tasks, delegates each to a specialised worker, inspects the results and decides the next action until the task is done.	Established
Handoff	A handoff transfers ownership of an ongoing task or conversation from one agent or service to another, together with the context it needs, so that the receiver continues the work directly.	Emerging
Human-in-the-loop	Human-in-the-loop pauses an automated workflow at a defined point until a person approves, rejects, edits or supplies input, then resumes with that decision as part of the workflow state.	Established
Event Choreography	Event choreography coordinates services without a central controller: each service reacts to events it subscribes to and publishes new events, and the overall workflow emerges from these reactions.	Established

Capability matrix

Which capabilities each pattern provides. Capabilities are the bridge between a requirement and a pattern.

Pattern	Persistent execution state	Resume after failure	Automatic retries	Fallback paths	Compensating actions	Parallel execution	Result aggregation	Dependency ordering	Runtime branching	Explicit states and transitions	Central delegation	Control transfer	Wait for external input	Decoupled coordination
Durable Execution	●	●	●	no	no	no	no	no	no	no	no	no	●	no
Saga / Compensation	●	no	●	no	●	no	no	no	no	no	no	no	no	no
Supervisor	no	no	no	no	no	no	●	no	●	no	●	no	no	no
Router	no	no	no	no	no	no	no	no	●	no	no	no	no	no
Handoff	no	no	no	no	no	no	no	no	●	no	no	●	no	no
Fan-out / Fan-in	no	no	no	no	no	●	●	no	no	no	no	no	no	no
Retry & Fallback	no	no	●	●	no	no	no	no	no	no	no	no	no	no
Human-in-the-loop	●	no	no	no	no	no	no	no	●	no	no	no	●	no
State Machine	●	no	no	no	no	no	no	no	●	●	no	no	no	no
DAG	no	no	●	no	no	●	no	●	no	no	no	no	no	no
Checkpointing	●	●	no	no	no	no	no	no	no	no	no	no	●	no
Event Choreography	no	no	no	no	no	●	no	no	no	no	no	no	no	●

Capability definitions

Persistent execution state: Progress, variables and position in the workflow are stored outside process memory, so a crash does not lose where execution was.
Resume after failure: A stopped or crashed run continues from its last recorded point instead of starting over.
Automatic retries: Failed steps are re-attempted according to a declared policy (attempts, backoff, retryable errors) rather than ad hoc code.
Fallback paths: When a step keeps failing, execution switches to a declared alternative (another provider, model, cache or degraded answer).
Compensating actions: Each completed step has a declared undo action that is run in reverse order when a later step fails.
Parallel execution: Independent units of work run concurrently instead of one after another.
Result aggregation: Outputs from concurrent branches are collected and merged at a defined join point.
Dependency ordering: Steps start only when the steps they depend on have finished, as declared in an explicit dependency graph.
Runtime branching: The next step is chosen at runtime from the current state, input or classification result.
Explicit states and transitions: The set of valid states and the allowed transitions between them are declared up front and enforced.
Central delegation: One coordinator assigns sub-tasks to specialised workers and decides what happens with their results.
Control transfer: Ownership of the conversation or task moves from one agent or service to another, together with the context it needs.
Wait for external input: Execution pauses until an outside signal arrives (an approval, a callback, a human edit) without holding compute while it waits.
Decoupled coordination: Participants coordinate by publishing and reacting to events, with no single component owning the end-to-end flow.