Rovo Grows Up: How Atlassian Turned Its AI Assistant Into an Agent That Does the Work

For two years, Atlassian’s Rovo was best understood as a very capable helper. It searched across your tools, answered questions in natural language, and drafted things on request. Useful, but fundamentally reactive, it waited for you to ask. At Team ’26 in Anaheim (May 5–7, 2026), that framing officially ended. The headline of the conference was the evolution of Rovo from an assistant into an agent that can plan and execute multi-step work on its own.

If you manage projects or ship software on Atlassian, this is the update that actually changes your week-to-week workflow. Here’s what landed, what it means, and where the sharp edges still are.

The foundation: an opened-up Teamwork Graph

None of the agentic features make sense without the Teamwork Graph, Atlassian’s context layer connecting people, work items, documents, code, and decisions across both Atlassian products and connected third-party tools. At Team ’26, Atlassian said the graph now holds more than 150 billion connections, and, more importantly, it opened that graph to outside tools.

Two interfaces went into open beta. The first is a Teamwork Graph command-line interface with more than 300 commands, which lets coding agents like Claude Code and Cursor query work and relationships across Atlassian products through a single interface instead of stitching together individual product APIs. The second is a set of Teamwork Graph tools delivered through Rovo’s Model Context Protocol (MCP) server. The strategic message is clear: Atlassian wants the graph to be the shared context that any agent, first-party or otherwise, reasons over.

Max mode: Rovo plans, then executes

The most consequential addition to day-to-day use is Max, a new reasoning mode for Rovo Chat, arriving in early access. Max is not just a smarter answer. It breaks a complex instruction into a multi-step plan, executes that plan across connected tools, and loops you back in only when a decision needs human oversight.

In practice, this is the difference between asking Rovo “what should I do about these duplicate bugs?” and telling it “find the duplicate defects in this project, merge them, reassign the survivors, and post a summary.” The second request is the kind of thing that used to mean a dozen back-and-forth prompts. Max is designed to handle it as a single instruction.

Agents in Jira reach general availability

Agents in Jira are now generally available, and this is where the agentic story becomes concrete for delivery teams. Agents no longer live in a separate chat sidebar. They live inside Jira with the same context your teammates have. You can assign work items to an agent, @mention one in a comment, or configure an agent to pick up work automatically when an item moves into a designated status.

Crucially for anyone responsible for governance, every agent action is logged in Jira with a full audit trail, and admins control which agents run and where. Atlassian’s framing, agents with the same accountability, visibility, and guardrails as the rest of the team, is the right framing to hold them to. You can connect first-party agents alongside third-party ones like GitHub Copilot, Cursor, Canva, Figma, and Gamma.

Rovo Studio goes GA: build agents without code

The other GA milestone is Rovo Studio, the no-code environment for building agents, automations, and apps grounded in the Teamwork Graph. The pitch is that you no longer need to know in advance whether your problem calls for an agent, an automation, or a full app, you describe the problem and Studio recommends and assembles the right combination. Apps (custom Forge apps generated from a single prompt) are still in beta; agents and automations are broadly available.

Atlassian backed this with adoption numbers worth noting: agentic automations have grown roughly sevenfold in six months, customers performed more than 14 million Rovo-assisted actions in the month before the conference, and the company says more than 90% of its enterprise cloud customers now use Rovo in some form. Customer results were specific…Mercedes-Benz reported agents cleaning up duplicate defects in test fleets and giving employees back a large share of their time, while Intermedia cited more than 50 hours of manual work saved per month.

The part you need to plan for: credits and governance

The capability is impressive; the operating model deserves attention before you roll it out. Rovo’s pricing is metered. Search is free and consumes no credits, but Rovo Chat and Agents cost credits per request, and heavier features like Deep Research cost substantially more. Credits are included in paid Standard, Premium, and Enterprise cloud plans, with the option to buy more. If you let agents run automatically on status transitions across busy projects, consumption is a real line item to model, not an afterthought.

Atlassian clearly anticipated this. New org-wide agent inventories give admins a live view of who built which agents, where they run, and how often. Permissions for AI access and for agent-building can now be separated, so you can let people use agents broadly while restricting who can create them, the main defense against agent sprawl. New dashboards and audit logs track adoption and credit consumption.

What to actually do with this

If you administer an Atlassian instance, three moves are worth making now. First, decide your agent-building policy before you open Rovo Studio widely: who can create agents, and where they’re allowed to act. Second, model expected credit consumption against the automations your teams realistically want, so the next renewal reflects actual usage rather than guesswork. Third, pilot Agents in Jira on a single team with a contained workflow (duplicate triage and status updates are good first candidates) and watch the audit trail before expanding.

The bigger picture is that Atlassian is positioning Rovo and the Teamwork Graph as the connective spine across Jira, Confluence, JSM, and your third-party tools…the layer that lets agents reason across all of it. Whether that vision holds up depends less on the demos and more on how cleanly the governance and cost controls work at scale. The capability has clearly arrived. The discipline to deploy it well is now the job.