Workflows
Deterministic processes — testable in dry-run, the platform's building-block promise.
A Workflow is a deterministic process: same input, same steps, same result — testable, verifiable in a dry run before it runs for real. Workflows are the building-block promise: predictable outcome, composable like bricks.
The rule that separates workflow from agent
“Can you write down the steps → workflow. Does it need case-by-case judgment → agent step, scoped and gated.”
A workflow is the right choice when you can describe the sequence in steps up front. The moment a step needs a case-by-case decision (is this reply a meeting request?), it belongs in a scoped, gated Agent step inside the workflow — not in a fixed rule that drifts away from reality.
The Pipeline: the standard workflow
Every Playbook runs on the Pipeline by default, the platform’s built-in standard workflow: sourcing → qualification → enrichment → personalization → sending. Every step has a defined prompt, a validated output schema, and the same learning-loop connection as every other step — deterministic in sequence, not in every individual result (an AI assessment stays an AI assessment).
What a workflow carries
- Trigger — what starts it: a new lead, a schedule, a manual kick-off through Command.
- Ordered steps — deterministic steps and, where needed, embedded agent steps.
- Approval point — wherever a step goes live and takes effect, Approvals apply.
- Mode — dry run shows what would happen, at no cost; live actually executes.
Every execution of a workflow leaves behind a Run — the receipt you read status, scope and cost from.