> ## Documentation Index
> Fetch the complete documentation index at: https://docs.e6data.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Build a workflow in chat

> Describe a repeatable outcome in plain English; AgentFlow plans, verifies and compiles the workflow before saving it.

export const DocMeta = ({type, applies, owner}) => {
  const LABELS = {
    tutorial: "Tutorial",
    "how-to": "How-to guide",
    explanation: "Explanation",
    reference: "Reference"
  };
  return <div className="e6-meta">
      <span className={type === "how-to" || type === "tutorial" ? "e6-eyebrow" : "e6-eyebrow e6-eyebrow--quiet"}>{LABELS[type] || type}</span>
      {applies ? <span>Applies to: {applies}</span> : null}
      {owner ? <span>Owner: {owner}</span> : null}
    </div>;
};

<DocMeta type="how-to" applies="All deployments" />

Workflows are authored in chat. You describe what you want, AgentFlow asks for missing details, proposes a plan, and waits for your confirmation before creating anything. The visual builder is for reviewing and editing once the first version exists.

## Before you describe it

Know these, so the clarifying questions go quickly:

* The repeatable outcome
* Runtime inputs
* The ordered operations and any branches
* Tools and skills each step needs
* Expected outputs
* What should happen on failure
* Any side effect that needs an [action policy](/agentflow/automate/actions-and-approvals)

## Create the workflow

1. Start or open a project chat.
2. Describe the workflow and its business rules.
3. Answer clarification questions about ambiguous steps, inputs, tool choice, branching, model needs and approval behavior.
4. Review the proposed plan.
5. Explicitly confirm that AgentFlow should create the workflow.
6. Wait while AgentFlow verifies and compiles it.

## What happens when you confirm

The authoring path:

1. Plans the graph and chooses step types.
2. Drafts the typed workflow specification.
3. Runs and verifies every deterministic `tool`, `transform` and `python` step.
4. Validates the full spec with the runtime compiler.
5. Repairs and rechecks any errors.
6. Saves only a workflow that compiles.

Creating and editing both apply the same compiler check when you save. Invalid specifications are refused.

## Related

* [Step types](/agentflow/reference/step-types)
* [Edit and version a workflow](/agentflow/automate/edit-and-version)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.