CEO and co-founder Mike Cannon-Brookes is betting Atlassian's Teamwork Graph can make agents visible participants across Jira, Confluence and developer workflows.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [Business Wire via Finviz](https://finviz.com/news/399638/atlassian-introduces-amp-the-agentic-multiplayer-protocol-to-power-human-ai-collaboration)
Why it matters #
Atlassian is betting its existing work graph and collaboration products can become the shared, permissioned workspace for human-agent teams. AMP's actual reach will depend on product adoption and whether its design patterns extend beyond Atlassian.
Atlassian introduced AMP, the Agentic Multiplayer Protocol, at its Team '26 Europe conference in Amsterdam on October 7th, proposing that AI agents work in shared team spaces rather than stay inside private chat windows and developer sessions. CEO and co-founder Mike Cannon-Brookes is making the pitch from Atlassian, which he built around software for teams: agents should be visible, identifiable and governed like participants in the work, not treated as a separate tool people consult alone. The announcement came during Atlassian's European conference, which runs October 6th through 8th.
Cannon-Brookes and Scott Farquhar met as students at the University of New South Wales and built Atlassian around software for teams. The company's history dates its business-name registration to 1998 and the launch of Jira 1.0 to 2001; Confluence 1.0 followed in 2002. Cannon-Brookes, who studied commerce and information systems at UNSW and became Atlassian's sole CEO in September 2024, is now arguing that the next layer of that work is coordination between people and agents.
A protocol built around the work graph
AMP's stated premise is that an agent's output should show up where a team already plans, discusses and reviews work. According to Atlassian, users can call agents through @mentions in Confluence, Jira comment threads and Loom briefs. Atlassian says its updated Atlassian MCP Server can connect those workflows to context in outside tools, including Figma and integrated development environments. Local agent sessions on developers' laptops can also be surfaced in Jira boards, making progress more visible to colleagues who were not in the terminal session.
The design extends to how agents appear and what they can do. Atlassian says each agent has an owner and a distinct profile, can appear alongside people in presence indicators and cursors, and can operate with a user's permissions or a dedicated service account. Atlassian says agent actions are recorded in an audit trail. Those controls address a real operational question for teams adopting agents: who authorized an action, what information could the agent access, and where can a colleague see what it changed?
Atlassian's AMP design material describes the offering as a collection of design patterns for human-agent collaboration, organized around real-time and asynchronous collaboration, identity and shared presence, and context and governance. That framing is more specific than the word "protocol" might suggest. The announcement describes capabilities across Atlassian products, while the public design material presents interaction patterns; it does not establish that AMP is an open technical standard that outside vendors can independently implement. For now, the clearest bet is that Atlassian can make its own collaboration suite the shared surface where human and agent work meets.
That bet leans on the Teamwork Graph, Atlassian's context layer connecting work and relationships across its products and connected tools. Atlassian says the graph now links more than 250 billion objects and relationships. The size of that graph is a company-reported scale measure, not evidence by itself that an agent's decisions improve. Its strategic value depends on whether the connections supply useful, permission-aware context and whether teams want agents acting inside the same systems where they already assign and track work.
Visibility is the product claim
The conference announcement bundled AMP with several related releases, including an updated Atlassian MCP Server, Agent Sessions, Non-Human Identities, Record for Agent and Interactive PR Reviews. Atlassian said its MCP Server had nearly 2 million monthly active users and more than 15 million daily tool calls, a 15-fold increase over six months. Those figures describe use of the MCP Server, not adoption of AMP itself. Atlassian also said its internal tests on Claude models found the updated server used up to 25% fewer tokens for the same Jira and Confluence work; that is an internal benchmark, not an independent productivity result.
MCP usage supports Atlassian's broader platform pitch. The company is presenting its established products and the Teamwork Graph as infrastructure for agent work, while the new capabilities make that work easier for other people to inspect and manage. Atlassian says people and agents already work together more than 10 million times a month across its platform, but it does not define that measure in the announcement. Atlassian also says it shipped more than 120 enterprise capabilities over the past year, a count that gives little indication of customer adoption or impact on its own.
For Cannon-Brookes, AMP extends Atlassian's founding proposition into a different kind of team. In the announcement, he said, "The best work has always been done in teams, and that's true of AI too." Whether agents become durable teammates will turn on more than putting a cursor next to a human's: teams will need dependable controls, clear accountability and measurable improvements to the work itself. AMP puts those requirements at the center of Atlassian's product design. The announcement establishes the direction; it does not yet show adoption beyond Atlassian's products or quantified gains for customers.