A Workflow agent is built as a visual flow rather than a form — you assemble the agent's behavior as connected nodes, so its behavior follows a defined, step-by-step process.
Overview
Note: Workflow agents are in limited preview and aren't available on most workspaces.
Instead of a configuration form, a Workflow agent gives you a canvas: you assemble the agent's behavior as connected nodes, and use Branch steps to split the flow into paths. Use a Workflow agent when the behavior is a defined process rather than open-ended conversation — where a Simple or Advanced agent reasons its way through a conversation, a Workflow agent follows the flow you laid out.
Choosing How It Starts
The empty canvas prompts you to Choose how AI Agent starts. You choose one of three triggers:
Trigger | Description |
|---|---|
Form | "Fill in a simple form to start the agent." |
Chat | "Talk directly to the agent in plain language." |
Schedule | "Run automatically at a scheduled time." |
Building the Flow
Once a trigger is chosen, you build the rest of the agent's behavior on the canvas. The add menu offers Add Node and Branch. The nodes you can add include:
Node | Description |
|---|---|
Prompt | Give the AI clear instructions to guide its thinking or behavior. |
Agent | Create a new mini-agent on the fly to handle a specific task. |
Search the Knowledge | Search across your company's internal knowledge. |
Respond | Craft the content, format and tone of the AI Agent's response. |
Actions from connected apps | Run an action in one of your connected apps. |
A Branch splits the flow into Path 1, Path 2 and so on, each with its own conditions, plus a Fallback Path for when no condition matches. Use Add Path to add more. This lets you route the flow differently depending on the situation, rather than relying on the model to decide at runtime the way a Simple or Advanced agent would.