Automations done by a person
Some steps of a workflow can't be done by an agent, such as setting up access in an external service or signing off a release. Make such a step an automation done by a person: when it starts, no agent runs. The run shows the automation's instructions and waits until someone does the work, writes a response, and submits it.
A person's automation has triggers and runs like any other, so it can sit in the middle of a chain. For example, an agent prepares a new integration, a person configures its connector and submits the connector ID, and the next agent automation continues with that ID.
How this differs from an agent's automation
Nothing runs in the cloud – the run starts no agent and no cloud environment, and it consumes no AI credits. The form hides the model, connectors, environment configuration, additional repositories, and each trigger's branch and change handling.
A repository is optional – the run checks out nothing. Select a repository only to show which one the step belongs to. In a chain, the next automation receives its name.
Every trigger works – a schedule, an event, a webhook, Run Now, or another automation finishing.
Nobody is notified when a run starts waiting – the notifications you add to the automation report a run only after it finishes, that is, after someone submits a response. Until then, the run is marked as waiting in the automation's Runs and on the Automations page, and that's the only place to find it.
A run waits until someone submits or stops it – there is no timeout. A scheduled person's automation adds a new waiting run on every tick, and the older ones keep waiting.
No templates – you can't save a person's automation as a template, and an automation linked to a template is always done by an agent.
Set up a person's step
Create an automation done by a person
Create an automation or open an existing one for editing.
In the form header, next to the scope and repository selectors, click Person.
If there is no Agent | Person switcher, automations done by a person aren't available in your organization yet.
In Instructions, describe what the person should do and what to write in the response. Instructions support Markdown.
Optionally, add triggers and notifications. Notifications report the run after someone submits it.
Save the automation.
On the Automations page, the automation gets a Person tag. Switching back to Agent restores the model and other agent settings you had before.
Do the step
Who can do the step depends on the automation's scope. Doing it is open to everyone who can see the automation, but rejecting it isn't:
Automation | Can submit a response | Can stop a waiting run |
|---|---|---|
Personal | Only you | Only you |
Project | Every project member | The automation's creator and project admins |
Complete a waiting run
Open the Automations page and select the automation in the Waiting state.
In the Runs section, click the run that shows Waiting for a person.
Read the Instructions and the Input. The input is what started the run: the trigger's event data, the context entered for a manual run, or, in a chain, the previous run's summary and links.
Do the work, then describe what you did in the Response field. The response supports Markdown and is optional: leave it empty to report that the step is done with nothing to add.
Click Submit.
The run finishes, and your response becomes its summary, with your name and the time you submitted it. A run can be submitted only once.
Reject a waiting run
If the step shouldn't be done, stop the run instead of submitting it. Project members who can't stop the run don't see Stop.
Open the run that shows Waiting for a person, as in the procedure above.
In the run header, click Stop.
In the Stop Run dialog, click Stop Run.
The run is cancelled and can't be submitted anymore. A cancelled run doesn't start the next automation in a chain. To ask for the step again, rerun it. A rerun uses the automation's current settings, so if it's still done by a person, the new run waits for a person again.
Pass a person's response to the next step
When a person submits a run, any automation that waits for it with an Automation finished trigger starts, the same as after an agent's run – see Set up a chain. The next automation receives the response as the previous run's summary. It receives no links and no pushed branch, because a person's run reports neither.
So write the instructions of the person's step with the next step in mind: ask for the IDs, URLs, or decisions the next automation needs, in a form its instructions can refer to.
A run keeps the instructions it started with. If you edit the automation while a run waits, even to switch it to Agent, the waiting run still asks for the original step and can still be submitted.