Atlassian introduced Agentic Multiplayer Protocol, a platform update that lets people and AI agents share context and tasks inside one workspace. Announced at Team '26 Europe, the move positions agents as governed participants with identity, authority and scope set by administrators. The company's MCP server already exposes 200 tools and processes about 15 million tool calls daily. The shift matters because agents move from isolated helpers to visible teammates working side by side with staff.
How Atlassian structures joint work of people and agents
Rovo, the company's AI assistant, anchors the new model by handling extended assignments that a person defines, reviews and approves. A new mode called Rovo Work inside Rovo Chat takes complex multi-step requests and can operate for hours, presenting a plan, accepting corrections and returning a result for final review. When a task falls outside its training, Work researches instructions and finds tooling, as in a case where it prepared an Instagram-ready reel by learning video synthesis. Power users can also have Rovo generate custom skills as part of its work, kept as a separately supported capability.
The protocol gives each agent the data needed to be useful while keeping it inside defined limits. Identity, authority and scope come from administration, so agents act within a governed system rather than connecting freely and operating without oversight. Head of AI Product Jamil Valliani described the setting as dynamic with many players, adding that autonomy without teamwork has little value. The internal line used by the company is "Headless software means brainless software." That design keeps human direction at the start, during execution and at approval.
Loom adds a visual instruction layer that replaces lengthy written prompts with demonstration. Users record themselves and their screen, circle interface elements, show clicks and field entries, and speak requests such as making an object larger or changing color. The recording captures speech together with on-screen references and formats them into a prompt for the agent. The approach mirrors how developers already guide coding agents and how two people align faster when looking at one screen. For routine operations, showing the desired outcome removes ambiguity that text alone often leaves.
What joint human-agent work means for companies
For operating teams, the practical gain is delegation of long-horizon work without losing control points. A manager can assign research, document drafting or preparation of materials in Confluence, where pages are live and structured, so humans and agents see edits in real time and adjust. Small firms get a way to run multi-step tasks without extra staff, while large organizations get traceability through identities and permissions. The difference from chatbots is continuity: objectives, plans and revisions stay inside shared tasks instead of scattered prompts.
For technology groups, the change expands the context available to agents beyond source code. Rovo Code Search brings repositories into the Teamwork Graph, a centralized engine mapping people, code and documents, while Data Context adds structured data from Databricks, Snowflake and BigQuery. The expanded Model Context Protocol server gives external coding and AI agents a common interface, with a substantial share of calls now writing information back rather than only reading it. Before selecting tools, check access scopes, data permissions, logging of agent actions and rules for publishing results.
The marker to watch is whether external agents routinely write finished work back into Atlassian through MCP at the current scale of 15 million daily tool calls. If write-back volumes grow and Rovo Work completes multi-hour assignments with human approval, shared workspaces will become the default handoff point. If activity stays read-only, agents will remain research assistants rather than teammates. The next product updates and customer cases after Team '26 Europe will show which path prevails.
