Atlassian's new protocol gives every AI agent an admin-assigned identity and scope

A user picks Rovo Work over ordinary Rovo Chat when sending a query. Jamil Valliani, Atlassian's head of AI product ^[4], described what that choice hands over. "When you opt to send a query to Rovo Work, you're actually giving Rovo permission to go and actually unleash itself fully," he said ^[8].

The pitch Atlassian made at Team '26 Europe ^[1] is that agents should work like employees inside a fully governed system of work ^[2]. "This is a multiplayer game," Valliani told SiliconANGLE. "Humans and agents are working together in a very dynamic space, and there are lots of players." ^[4] Each agent gets an identity assigned by administration, with authority and scope, plus limits meant to keep it inside the platform ^[3]. An administrator sets those before any user types a request.

Teams count on the user for control: the user sets the objective, reviews the proposed plan, course-corrects and reviews the result ^[7]. But users hand Rovo a long task, and it can disappear for hours ^[7]. When the task is new to it, Rovo Work goes and finds what it needs, tooling included ^[11]. The human checkpoints sit at either end. For the hours in between, the limit is whatever scope the admin wrote ^[3].

That scope now covers more ground. With source code added to the Teamwork Graph ^[10], admins writing agent scope against the graph have repositories to account for as well as documents.

Atlassian says openly that it cannot predict what people will ask for. "We can't really imagine the creativity our customers are going to ask it to unleash," Valliani said. "We want customers to go and challenge it, and tell it what they really want." ^[13] SiliconANGLE's report says agents cannot run rampant ^[9]. Valliani gave the company's internal statement as "Headless software means brainless software," and its position is that autonomy is not useful without teamwork ^[5].

The report does not describe the screen where identity and scope are set, the default scope a new agent starts with, or pricing and availability.

Which agents need setup first depends on how long each one works between human checkpoints and what it can reach. On duration, it might give a single chat answer or run a Rovo Work task for hours ^[6]. On reach, it might use context already in the platform, or tools and skills it picks up on its own; Rovo can generate custom skills for power users ^[12]. For a short task on platform data, the user reading the answer is enough review. A long task on platform data depends on the scope matching the project. When a short task pulls in a new tool, the user who asked sees it in the output. In my view, the fourth box needs the most care: long tasks where the agent gets its own tooling. For those, identity, authority and scope should be written down before the first query, along with the name of the person who answers for the result.