Workflow Automation runs a defined sequence of steps: check something, produce something, wait for a person, then continue. Every run is recorded. ## Designing a workflow Describe the outcome in plain English and use **Design with AI** to get a first draft, or add steps yourself. Each step has a type, a description and, where relevant, conditions. Conditions can depend on the day of the week, the hour, a previous step's output or a value you provide. A workflow built for "every weekday morning" ends up with an hourly and a weekday condition rather than one vague instruction. ## Approvals and pause points A step can be marked as needing approval. The run stops at that step, the request appears in your approvals inbox, and nothing after it happens until you approve or reject it. Rejecting ends that branch of the run. This is how a workflow can update one of your support tickets: the ticket change is prepared, you approve it, then it is applied. Steps that send a signed webhook to an endpoint you configured behave the same way — on plans that include integrations. ## Running Run a workflow on demand, or let the scheduler start it. Each run has a status: completed, cancelled, partial or waiting for approval. A partial run means some steps succeeded and others did not; the run detail shows which. Workflow runs create tasks in your AI Task Manager rather than silently changing records, unless a step is explicitly an approved ticket update or an approved webhook. ## Cancelling You can cancel a run that is queued or waiting. Cancelling does not undo work already done, and the history says what had completed.