Atlassian opens Teamwork Graph to third-party agents at Team '26

Atlassian on Wednesday opened its Teamwork Graph data layer to third-party AI agents and rolled an expanded suite of Rovo products into general availability at its Team '26 conference in Anaheim, California, in a push to position the Sydney-headquartered software maker’s enterprise context as the spine for agentic work.

Co-founder and chief executive Mike Cannon-Brookes told an audience of customers and partners that frontier AI models were converging on quality, and that competitive advantage now depended on connecting agents to a company’s accumulated history of decisions, tickets and code. “In 2026, anyone can buy ‘smarts’ by the token,” Cannon-Brookes said in his Team '26 founder update. “The real moat is your institutional memory.”

The company said Rovo, the AI agent platform that plugs into Atlassian’s Jira, Confluence and Bitbucket products, is now used by 75 per cent of Fortune 500 companies and 90 per cent of Atlassian enterprise cloud customers. Customers ran more than 14 million Rovo-assisted actions in the past month, according to the company, with customer-built agentic automations rising sevenfold over six months.

The Teamwork Graph itself, the data layer that maps how work moves across Jira, Confluence, code repositories and connected SaaS tools, now contains more than 150 billion connections, Atlassian said.

Opening the graph to outside agents

The headline announcement was an open beta of a Teamwork Graph MCP server, which exposes graph context to external AI agents through Anthropic’s Model Context Protocol, alongside a Teamwork Graph CLI offering more than 300 graph-aware commands for developer-side AI tools such as Claude Code, GitHub Copilot and Cursor. Atlassian cast the move as a concession that no single vendor’s agents will handle every workflow inside a customer.

Figma and Replit were named as launch integration partners for the open graph. Atlassian also showed Rovo Chat pulling 20 years of customer history from 61 connected sources across Salesforce, Jira and Confluence inside three minutes, generating dynamic charts and a stakeholder map, in a stage demo led by Atlassian executive Sherif Mansour.

“Acceleration for your business is about context multiplied by intelligence,” Cannon-Brookes said.

A no-code agent builder and Jira as control plane

Rovo Studio, the no-code interface for non-technical staff to build agents against company data, moved to general availability at the conference. Atlassian also launched Agents in Jira, which lets organisations assign tickets to agents the same way they assign work to people, and opened a beta of Remix with Rovo inside Confluence for repurposing existing pages.

A higher-reasoning Max mode in Rovo Chat is in coming-soon status. The company also opened early access to Code Intelligence, a semantic search layer that spans 11 million files and 1.5 billion lines of code in test deployments. In a stage demo, the tool surfaced TODO comments more than two decades old still embedded in a customer codebase.

The DX AI Experience product, which measures the return on investment of AI rollouts inside engineering teams, is now generally available. Atlassian said it had also expanded availability of Dia, the AI browser assistant it picked up through last year’s acquisition of The Browser Company.

In a separate stage demo, Mansour showed Jira routing a single bug ticket between a human engineer, an automated workflow, and an external agent running in Anthropic’s Claude Code, with the graph keeping an audit trail of which step each component completed. Executives said the company expected most enterprise customers to run a mix of vendor agents rather than commit to one platform.

Selling the nervous system, not a clean slate

Industry commentary after the keynote cast the announcements as Atlassian selling its installed base “the nervous system” for agent-driven work rather than asking customers to switch platforms. Mansour leaned into that on stage with examples of decade-old technical debt that the graph could surface for review.

Cannon-Brookes acknowledged the discomfort that came with the new visibility. “Work will always be a little bit messy,” he said. “That’s where the human ingenuity actually lives.” He told the audience that opting out of the agentic transition was not a viable strategy. “You cannot wait and see your way through an existential shift in technology.”

The chief executive also signalled a shift in how the company would release product. Atlassian would move away from periodic flagship updates toward “building and shipping in public, every day,” he said.

Shares in Atlassian, dual-listed on Nasdaq, traded at $US92.37 around the time of the keynote, against a consensus analyst price target of $US142.54, according to Yahoo Finance data published the same week.

The Team '26 announcements landed amid a string of AI moves by Australian software firms. Sydney design platform Canva paused 5,300 staff for an internal AI Discovery Week with OpenAI, Anthropic and Google last week, while voice-AI maker ElevenLabs opened a Sydney office in the same period and named a local general manager.