Atlassian Rovo

In Rovo chat, Atlassian’s tools guide says the agent asks for confirmation before running consequential tools that may change data. A person asks for a Jira work item, the agent asks for any missing required fields, and the person confirms before the work item exists.

Put the agent in an Atlassian Automation rule and the person drops out. Atlassian’s automation page says agents in automations can be configured without confirming every action, and the Use agent action’s configuration guide says agents in automations can now perform write actions directly. The general tools page still says an automation-triggered agent can’t use its own tools and only returns text. The docs disagree, so the tenant’s actual settings decide.

Removing the confirmation is only the visible change. Three other things can move with the job:

• Whose permissions run the action. In Studio, an agent can act as the person using the agent or as the agent’s own account. With a person’s account in an automation, actions are logged against whoever set up the automation.
• Where the result lands. In Atlassian’s recorded warranty audit, a scheduled rule runs the agent, and a separate step creates a Jira task with the agent’s response as the description.
• Who notices a mistake. In chat, the requester is present before the change. In an automation, the builder chose the schedule and nobody watches each run.

Atlassian’s safety guide recommends narrowing an automated agent’s tools and, where it fits, duplicating an agent for one automated task. For builders: moving a job from chat to an automation is a new permission review. Check the acting account, the allowed writes and the destination before the first unattended run.