Building apps
What is an AI workflow?
An AI workflow is a defined sequence of steps in which an AI model performs part of the work, with the surrounding steps handling input, validation, and what happens next.
New to Bolt? Sign up free.
Technical definition
An AI workflow is an orchestrated pipeline combining deterministic steps with one or more model inference steps. A typical structure is: a trigger, retrieval or preparation of the input, a prompt constructed from that input, inference, validation of the output, and a downstream action or storage step.
Workflows differ from agents in control flow. A workflow follows a path you defined, with the model filling specified slots. An agent decides its own sequence of actions toward a goal.
In plain terms
An AI workflow is a process where one or more steps are done by AI, and you decided the order.
An email arrives. The workflow pulls out the details, has AI draft a reply, shows it to you for approval, then sends it and files the record.
Why it matters
Most useful AI in a small business is a workflow rather than an agent, and the reason is control.
A workflow is predictable. You know which steps run, in what order, and where the model is involved. When something goes wrong you can find the step that did it. An agent choosing its own path is more capable and much harder to debug, which is why the workflow is the right starting shape for anything touching customers or money.
Designing one comes down to four decisions:
- Where the model sits. Only the steps needing interpretation.
- What it sees. The narrowest input that lets it do the job.
- How its output is checked. Format validation at minimum, human approval for anything consequential.
- What happens when it fails. Not an unhandled error, and not silently proceeding.
The most common failure is the fourth one being unspecified. The model returns something unexpected, nothing catches it, and the workflow either stops without telling anyone or passes bad data downstream.
