Rovo Agent permissions and governance

Rovo enables everyone in your organization to create Agents.

Just like creating a page to share an idea or thought, teams should be able to uncover insights, brainstorm, move work forward, or save time by creating an Agent to assist with something.

We appreciate this may raise some questions about how you use Agents in your organization.

What are Agents allowed to do?

Rovo Agents can only do what the user can do, not the creator. Using an Agent never gives anyone more permissions.

So the creator can create an Agent, but the Agent always respects the user's permissions:

• If the user doesn’t have access to view a page, the Agent can’t respond with anything about that page.

• If the user can’t comment, the Agent can’t comment.

• If the user can’t delete a Confluence page, the Agent can’t delete the Confluence page.

An agent also can’t access information for a user that they wouldn’t already have access to. For example, a user can’t access someone else’s private Confluence page through an Agent, even if they created the Agent.

Can an Agent work autonomously

For Agents to work autonomously, they must be managed through your existing Confluence space or Jira Project administrators via an Automation rule.

Without Automations and setup from an Admin, Agents can’t work autonomously.

More on Agents and automations

Will creating many Agents impact performance?

There is no negative impact on system performance or scale based on the number of Agents created.

How will users find the right Agents for their work?

You can search and find Agents using the Browse Agents view. More Agent categorization features are coming soon, including natural language search and more contextually relevant recommendations.

What if someone creates a ‘bad’ Agent?

Ethical filtering setup is across all our AI to protect your teammates from anything inappropriate. This applies to Chat, Agents, and all our other generative AI.

Because of these filters, ‘bad’ generally only extends to output that isn’t very helpful, in which case, we encourage people to keep editing and tweaking the Agent to improve its output. Learn more about instructing an Agent

Can I control what an Agent can and can’t say?

Generative AI is probabilistic by nature which means you can’t ever make the output 100% pre-determined. Through experimentation with your Agent’s instructions, you can get close to a deterministic outcome. More tips and best practices for writing instructions

Is there an audit log for Agents?

There is currently no dedicated audit log, but every Agent profile lists every autonomous action that Agent has performed.

If you’re concerned about your Agent doing things outside of its intended purpose, try including those things in your instructions. Agent instruction can be improved by specifying what you DON’T wan’t your Agent to do.

You can also adjust the list of actions that your Agent has access to. For example, if you no longer want your Agent to create Confluence pages, you can remove that action by editing your Agent. More on editing your Agent.