Atlassian creates the basis for collaboration between humans and AI agents

Atlassian Corp. today announced the Agentic Multiplayer Protocol, a platform update designed to transform the way people and artificial intelligence agents collaborate and get their work done using shared contexts and tasks in the same digital space.

At the unveiling at Team ’26 Europe, Atlassian said it wants AI agents to function like employees, bringing what it calls “multiplayer” into a fully managed work system so that humans and agents can operate side by side in the open.

AMP is designed to give each agent the data context it needs to be useful and constraints to remain secure within the boundaries of the platform – this includes a management-assigned identity, including authority and scope.

“This is a multiplayer game,” AI product lead Jamil Valliani said in an interview with SiliconANGLE. “People and agents work together in a very dynamic space, and there are many actors.”

However, agents cannot operate rampantly. Valliani added that Atlassian has an internal statement: “Headless software means brainless software.” The company’s mission is not just to provide software for autonomous agents that can connect with and work without humans; Autonomy is useful, but without teamwork it is not useful.

Rovo, the company’s AI assistant turned autonomous powerhouse, can disappear for hours to work, but the user sets the goal and guides it, reviews the proposed plan, italicizes it, and finally checks the result.

To address this issue, Atlassian announced Rovo Work, a new mode for Rovo Chat that handles complex, multi-step tasks that are reviewed and approved by humans. It is designed for tasks with a long time horizon.

“When you choose to submit a request to Rovo Work, you are actually giving Rovo permission to fully develop,” Valliani said.

In fact, Rovo can go a step further than simply completing work. For example, if Work is asked to approach a task it is unfamiliar with and it doesn’t have what it needs in its model training data, it will attempt to train itself.

In one case, a product manager asked Work for an Instagram-ready Reel. Instead of using everything that came with its base model, Rovo researched the proper instructions, learned what it needed for the output format, and got tools for video synthesis.

This is significantly different from the standard AI modeling approach of “Is this already available in my training?” to “How do I learn this?”

Valliani also confirmed that Rovo could generate custom skills for power users as part of its work. Although this is a separately supported feature, not every learned behavior becomes a skill that users can export.

“We can’t really imagine the creativity our customers will unleash with this,” Valliani said. “We want customers to rise to the challenge and tell her what they really want.”

Loom: Replace prompt engineering with “Show me what you mean.”

Atlassian also brings something to everyday users that developers have long used with agents when coding: showing instead of telling.

Loom is a product made by Atlassian that allows users to record videos of themselves and their computer screen and share them with each other to provide instruction. It can also be used to train agents. For example, Loom can be used to draw circles around parts of the interface, show button clicks, and enter information into fields.

The problem with AI agents is not always intelligence. Sometimes people find it difficult to explain what they want with words, but if they can use words to visually show what they need: “Make this bigger, move that, make it green.” This makes the entire process trivial. Two people looking at the same screen while talking to each other have the same context and they have a visual “prompt” plus the words.

“With Loom, it actually captures how you say all of this and the actual references on the screen that you point to when you say it and then formats it into the prompt,” Valliani said.

Atlassian expands the framework around code and context for developers

For developers, Atlassian’s broader platform changes mean expanded context and operating environment for AI agents.

Introduced today, Rovo Code Search brings the source itself into the Teamwork Graph, a centralized data intelligence and context engine that maps relationships between people, code and documents. Data Context extends this view to structured information stored on platforms such as Databricks Inc., Snowflake Inc. and Google LLC’s BigQuery. Together, they give agents a more comprehensive view of not only how software is built, but also the business information surrounding why software is built.

The company’s advanced Model Context Protocol server also provides external coding and AI agents with a common interface to its platform. Valliani said a significant portion of these interactions now involve agents writing information back, not just retrieving it. MCP transforms Atlassian from a data source into a part of the shared workspace, where agents leave the work for humans and other agents to take over.

Atlassian’s MCP server now provides 200 tools and processes around 15 million tool calls every day.

This is reminiscent of the company’s desire to make agents part of the team and the “multiplayer” formulation that gives agents a place in the company, roles and permissions.

Imagine an AI agent working in Confluence, the company’s digital knowledge management and document collaboration workspace that resembles an internal central wiki, seeing another agent or human editing the same document. Because all pages are live and structured, the company knows in real time that it is collaborating with another teammate, checks for changes, and adjusts its approach.

Image: SiliconANGLE/Microsoft Designer