A workflow is the map of how work flows through a process. It has a starting event, a series of steps, often with decisions and branches (if this, then that), and an end state. It can be carried out by people, by software or by both.
In automation, a workflow is the thing you build and run: a webhook receives an order, a step checks stock through an API, a branch handles the out-of-stock case, and a final step emails the customer. Each run is an execution of that workflow, with its own inputs and result.
Good workflows are small, named for what they do, and designed for failure: what happens if a step errors, times out or runs twice? Answering that is what error handling, retry logic and idempotency are for. Used with AI models, a fixed workflow is more predictable than an open-ended AI agent.
Related: Workflow Automation, n8n, AI Automation.