How do complex workflows delegate tasks to other agents?

A workflow agent can hand off a bounded sub-task to a specialist sub-agent, wait for it to complete, and use the results to continue—enabling sophisticated multi-step automation without losing oversight.

For complex processes, a main workflow agent can hand off a bounded task to a sub-agent with its own specialist skill set, then wait for the result before continuing.

Why this matters

The main agent stays focused on orchestrating the overall process.
The sub-agent handles a specialised task—such as gathering information from a third party or running a tool-heavy investigation.
Neither agent is overloaded with skills it does not need, which keeps them focused and reduces mistakes.

How delegation works

You configure the main agent step with a set of delegatable skills (separate from its own skills).
When the main agent encounters a step that needs one of those skills, it hands off a bounded task to a sub-agent with specific instructions.
The main workflow pauses.
The sub-agent runs, potentially pausing itself for outreach or waiting for a reply.
When the sub-agent finishes, its result and any files are passed back to the main agent.
The main agent continues with full context of what the sub-agent found.

One delegation at a time

Only one delegated sub-agent runs for the parent at a time: the main workflow pauses until that sub-agent finishes (or times out) before the parent can delegate again. You cannot “fan out” to two sub-agents in parallel in a single turn.

If you need two different people each to have a separate back-and-forth conversation, the main agent will typically contact them in sequence—first delegation completes, then the parent resumes and can delegate to the next. Plan for longer wall-clock time and set timeouts accordingly.

For choosing between one configurable agent step and Advanced mode (canvas editor)—and how delegation fits—see When should I use a single-step agent workflow instead of the canvas editor?

A common pattern: gather then act

A frequently used delegation pattern:

Main agent: orchestrates the workflow, drafts documents, updates tickets.
Sub-agent: contacts an employee or manager (via Slack or email) to gather a specific piece of information, then hands back a clean summary.

This keeps the information-gathering conversation separate from the main workflow, ensures the data-gathering agent cannot accidentally access broader HR data, and gives you a clear audit trail for both parts.

Example

A leave policy workflow needs to verify that the employee's manager has confirmed a return date.

Main agent delegates to a data-gathering sub-agent: "Contact {manager_email} via Slack and confirm {employee_name}'s agreed return date."
Main workflow pauses.
Sub-agent messages the manager, waits for their reply, and confirms the date.
Main workflow resumes with return_date: 2025-09-03, confirmed_by: manager@company.com.
Main agent generates the return-to-work letter and closes the task.

Guardrails

Sub-agents cannot delegate further—delegation is one level deep by design.
Sub-agents only see the skills explicitly listed as delegatable—they cannot access the main agent's full toolset.
Set a timeout on sub-agent steps so the main workflow does not wait indefinitely. If a sub-agent times out, the main agent's on error instructions should specify what to do next.
If you need the sub-agent to pass files back (for example, a completed form), it can—files are automatically passed to the parent when the sub-agent completes.