Atlassian has introduced a protocol meant to let people and AI agents work in the same digital spaces, along with a new mode in its Rovo assistant for long-running tasks, SiliconANGLE reported Wednesday. The announcements came at the company's Team '26 Europe conference and also covered screen-recording instructions, code search and links to outside data warehouses.

What Happened

The centerpiece is the Agentic Multiplayer Protocol, or AMP. Under it, each agent working in an Atlassian space gets the data context it needs, plus an identity, a level of authority and a defined scope that an administrator assigns. The aim is for agents to operate alongside employees in shared projects rather than in separate chat windows.

Rovo Work is a new mode of Rovo Chat built for complex, multi-step jobs that may run for a long time. A user sets the goal, reviews the plan the agent proposes and approves the results before anything is carried out. Atlassian said agents in this mode can research tasks they have not seen before. In one example described by SiliconANGLE, Rovo worked out how to make an Instagram reel, learned the format requirements and found video tools to produce it.

RelatedClaude Code Desktop Preview Flaw Could Have Exposed Developers’ Cloud Keys, Researchers Say

Atlassian also added a Loom integration that lets people instruct agents by recording their screen, pointing at buttons and fields and describing changes out loud instead of typing a prompt. Rovo Code Search pulls source code into the Teamwork Graph, Atlassian's map of how people, documents and code relate. A feature called Data Context extends agents' reach to structured data stored in Databricks, Snowflake and Google BigQuery.

The company's Model Context Protocol server now exposes about 200 tools and handles roughly 15 million tool calls a day, according to SiliconANGLE. It also lets outside agents write information back into Atlassian products, not just read from them.

Why It Matters

Atlassian's pitch is that agents need the context of how a team works, not just access to an application's data. Jamil Valliani, Atlassian's head of AI product, framed the argument bluntly, saying "headless software means brainless software." That is a direct response to the idea that agents will bypass collaboration apps and talk straight to back-end systems.

The identity and scope controls in AMP address a concern that security and IT teams raise about agents: who is accountable when software acts on its own inside business systems. By making administrators assign each agent a role, Atlassian is treating agents more like new team members than like plug-ins.

The MCP figures matter too. MCP, the open protocol Anthropic introduced in 2024, has become a common way for agents to call business software, and 15 million daily calls suggests outside agents already lean heavily on Jira and Confluence data.

Context

The announcements follow Atlassian's disclosure this week that it has committed spending to OpenAI while keeping Rovo able to route work across several model providers. The same week, Atlassian also warned customers about a critical file-read flaw in its self-hosted Data Center products.

What to Watch

Atlassian did not give pricing or general availability dates for AMP or Rovo Work. Key questions are how the protocol's permission model holds up when agents from other vendors join Atlassian spaces, whether the Databricks, Snowflake and BigQuery links carry extra charges, and how quickly customers move long-running work into Rovo Work once it ships.