Agent Task

Task type: Agent

Use Agent Task when a deterministic workflow needs a flexible AI step in the middle. The workflow stays in control of when the step runs, and the agent handles interpretation, tool choice, or synthesis.

Run behavior

• Side effects: This task is read-only or computation-only by default. It returns data for later steps rather than directly writing records or sending messages.
• Credit usage: This task can consume organization credits or model/research capacity depending on your plan and configuration.

When to use this task

• Interpret a transcript or research result before choosing the next workflow branch.
• Ask a specialized agent to summarize context or draft a recommendation.
• Reuse an existing custom agent inside a larger workflow process.

Setup tips

• Create and test the agent first, then link it from the workflow task.
• Pass only the inputs the agent needs instead of the entire workflow state.
• Keep the prompt clear about what the agent should return to the workflow.

Examples

Meeting follow-up router

A workflow reads a meeting transcript, then calls an agent to decide whether the account needs a support escalation, executive follow-up, or standard nurture.

Outputs

• Agent response text or structured output, depending on the linked agent configuration.

Common pitfalls

• Using an agent when a simple workflow branch would be more predictable.
• Passing too much context and making the agent harder to test.
• Forgetting that model output can vary even inside a controlled workflow graph.

Related guides

• Build a multi-tool agent
• Workflows vs. agents

Reference details

These details are generated from the workflow task registry and match the fields available in the builder.

Required fields

• agentId
• prompt

Variable-aware fields

Fields that accept variable references ({{ inputs.x }}, {{ steps.N.y }}) from the workflow's variable graph:

• inputs

Liquid template fields

Fields that are rendered through Liquid (variables plus filters and control flow):

• prompt

Model guidance

Delegate agent authoring to the Agent Builder Genie, then link the saved agent id.

Builder guidance

When the Workflow Builder Genie configures this task, it follows this guidance:

Configure the agent linkage only. If a new agent is needed, create/open it in the Nested Agent tab and let the Agent Builder Genie write the agent prompts and tools.

Default configuration

New instances of this task start with the following configuration:

{
  "name": "Agent Task",
  "outputKey": "agent_task",
  "taskType": "Agent",
  "description": "",
  "agentId": "",
  "prompt": "",
  "inputs": {}
}