Atlassian unveils Agentic Multiplayer Protocol for AI agents

Atlassian framed the announcement around an internal saying from Head of AI Product Jamil Valliani: “Headless software means brainless software.” The point is that autonomy alone is not the goal. The company wants agents to sit inside a governed system where they have an identity, limits and a place on the team.

That thinking shows up across the products it introduced or expanded. Rovo Work gives an agent permission to take on long, multi-step tasks, but a person sets the objective, reviews the plan and approves the result. Loom lets users show rather than tell by recording their screen and voice, which Atlassian says becomes a visual prompt. For developers, Rovo Code Search and Data Context pull code and business data into the Teamwork Graph, while an expanded Model Context Protocol server lets external agents read from and write back to Atlassian’s platform.

The bet is that agents become teammates rather than tools you query. Whether that holds may depend on whether the governance Atlassian describes — admin-assigned identity, authority and scope — actually keeps agents useful without letting them drift. The example the company gives is an agent in Confluence noticing another editor on the same live document and adjusting its approach, which suggests the value hinges on real-time awareness as much as raw capability.