SubAgent Spawning¶

Some workflows benefit from delegating a complex sub-task to an Agent step. Conceptually, an Agent step is “a workflow inside a workflow”: it can take an objective, do focused work, and return a result back to the main flow.

This is useful for tasks like research, multi-part analysis, or producing structured outputs that then feed into downstream steps.

When To Use An Agent Step¶

Agent steps are a good fit when you need:

A focused sub-task with a clear deliverable (summary, table, classification, plan)
Multi-perspective output (for example: “give me 3 viewpoints” or “compare options”)
A sub-process that benefits from deeper exploration before producing an answer

If the task is simple (format a value, filter a list), prefer a code step or a direct integration step instead.

Configuring Agent Steps¶

In the workflow editor’s node drawer, Agent steps commonly support:

An objective (what the sub-agent should accomplish)
Briefing/context (what it should consider, and what format to return)
Optional constraints (for example: which tools/services it is allowed to use)
Optional parallelization (multiple “children” working on different angles)

Virtual Run will preview Agent steps so you can see what they would produce before you run the workflow live.