Skip to main content
Your coworker uses the automations its agent already has: it starts a workflow, waits for the result, and carries on.

What it does

You’ve probably already automated some of your work with workflows: an onboarding workflow, a nightly sync. Your coworker can use the ones in its own agent. Work in other agents it reaches only through what those agents offer, their endpoints. Ask for something in plain English, like “run the onboarding workflow for Northwind”, and it finds the right one and starts it. It waits for the run and reads what came out. It can also read its agent’s past runs, and the conversations its conversation flows have had. It doesn’t build or change automations. Build them yourself, or ask the Concierge.

When you’d use it

  • “Run the onboarding workflow for this new customer”
  • “Did last night’s sync actually run?”
  • “What did the sales chat tell this lead yesterday?”
  • Kick off an automation as the last step of a longer job
  • Check the result of a run before deciding what to do next

Turning it on

There’s nothing to turn on. Every coworker can run its agent’s workflows, so its Capabilities tab doesn’t list it. You can switch it off for one channel, under Capabilities on this channel in the channel’s settings, where it’s called The agent’s own tools. See Channel settings.
The Manage workflows and flows row on the Capabilities tab, switched on

Reading and running are free; changes wait for your approval.

Reading versus running

The distinction is worth holding onto, because it’s the whole safety story of this capability. When it wants to start a workflow, you get a card in the conversation naming it. Approve it and the run starts; decline and it doesn’t, and your coworker carries on with the rest of the job. Always allow on that card stops the asking for this coworker. A run it starts is waited on for up to two minutes. If it takes longer, your coworker gets the run’s id and checks back on it.

Good to know

This is the fastest way to get value out of automations you already have. A coworker that can start your existing onboarding workflow is useful on day one, with nothing new to build.
You can make this stricter or looser with Permission rules, for example letting a trusted coworker run automations without asking.

Common questions

No. It runs workflows but never changes them, and each run asks you first until you choose Always allow.
It reads what’s in its agent and matches on the name and what the workflow does. If more than one could fit, it asks rather than guessing.
Only one that agent offers to it, as an endpoint. It calls that without asking you first, unless a permission rule says otherwise.
A workflow runs steps in order. A conversation flow holds a conversation with a customer. Your coworker runs its agent’s workflows and reads its conversation flows’ conversations. See Cowork vs Workflows vs Conversation flows.
Yes. It can check on a run and read what came out, then use that in the rest of the session.

Next steps

Approvals

How a run request reaches you.

Permission rules

Loosen or tighten what needs asking.

Building a workflow

The automations your coworker runs.

Cowork vs workflows

Which tool fits which job.