Why not put the whole workflow in one Lambda function?

One function can be appropriate for short, cohesive work. A managed workflow becomes valuable when state, waits, branches, retries, auditability, or partial recovery need to remain visible outside one process.

Are workflows an alternative to events?

No. Events can start workflows, workflows can publish events, and tasks can wait for callbacks. Use each for the coordination responsibility it expresses clearly.

Is AWS Step Functions a serverless workflow service?

Yes. Step Functions is a managed workflow service based on state machines and tasks for distributed applications, automation, microservices, and data or machine-learning pipelines.

What are serverless workflows?

Learn when explicit workflow state improves retries, branching, parallel work, human approval, visibility, and recovery in distributed applications.

What are serverless workflows?

A serverless workflow coordinates a sequence of tasks with a managed service that stores execution state, applies transitions, and handles control flow without requiring the application team to operate a workflow server fleet.

How this topic fits

Workflows

Coordinate multi-step work with visible state, retries, timeouts, compensation, approvals, and long-running business processes.

Why distributed work needs coordination

A business operation often spans validation, data changes, external APIs, notifications, and recovery. Embedding the entire sequence in one function can hide state and make partial failure difficult to inspect or resume.

A workflow makes steps, branches, waits, retries, and terminal outcomes explicit. The execution history becomes an operational artifact that shows what ran, what failed, and what input reached each decision.

Run tasks in a defined order and pass only the data each step needs.

Branch, wait, repeat, or run tasks in parallel based on explicit state and rules.

Retry transient failures, catch expected errors, compensate changes, or resume from a reviewed point.

Orchestration and choreography

Orchestration uses a central workflow definition to decide which task runs next. It provides a visible control path and is useful when the end-to-end operation needs one owner or audit trail.

Choreography lets services react to events without a central coordinator. It can reduce central coupling, but the complete business path becomes harder to see. A system can use orchestration within one process and events between independently owned processes.

Design task and state contracts

Give each task a bounded responsibility, typed input, explicit output, timeout, and error taxonomy. Avoid passing the full workflow document through every step when a stable identifier and narrow state are enough.

Decide which state belongs in the workflow history and which belongs in a durable domain store. Protect sensitive data, constrain execution history access, and plan how schema changes affect executions already in progress.

Long-running work and human approval

Managed workflows can pause for a timer, external callback, or human decision without holding a compute process open. That makes them useful for approvals, fulfillment, data pipelines, and integrations with uncertain completion times.

Define expiration, escalation, cancellation, duplicate callback, and operator-recovery behavior. A workflow that waits safely still needs a clear owner and an alarm when the expected external action never arrives.