Atlassian Unveils AMP to Govern Human Teams and AI Agents

On October 7, 2026 Atlassian officially announced the launch of the Agentic Multiplayer Protocol, commonly referred to as AMP. The initiative was presented as a direct response to the growing demand from organizations that need to create, operate, and supervise mixed teams composed of both human personnel and artificial‑intelligence agents. The announcement took place at the company’s annual conference, an event that brings together customers, partners, and developers to showcase Atlassian’s latest innovations. Executives emphasized that as AI agents become active collaborators in critical workflow pipelines, providing comprehensive visibility and granular control over their actions becomes essential for ensuring compliance, security, and accountability across enterprise environments.

What Is AMP and How Does Atlassian Define It?

According to Jamil Valliani, Atlassian’s Vice President of AI, AMP is not a formal standard issued by an external standards body; rather, it is an integrated suite of technologies, user‑interface patterns, and governance mechanisms. The core idea behind AMP is to deliver a software layer that enables administrators to treat AI agents as full‑fledged team members, complete with clearly defined rights and responsibilities. This approach eliminates the need for organizations to adopt multiple external protocols or to craft bespoke rules for each AI provider, thereby simplifying management and ensuring consistent compliance across the organization.

Teamwork Graph‑Based Architecture

AMP relies on the Teamwork Graph infrastructure, the relational data engine Atlassian uses to share context among people, projects, and content. Through this graph, contextual information propagates with fine‑grained permissions and is recorded in detailed audit logs. Every entity—whether a document, a task, or an agent—is linked to a node in the graph, ensuring that permissions and access histories are applied consistently throughout the entire Atlassian product ecosystem, including Confluence, Jira, and future extensions.

Agent Governance and Admin Controls

In the administration console, AI agents appear as distinct accounts that are visible and manageable by administrators. From this unified view, responsible parties can review every action performed by an agent, suspend its account, adjust its permissions, or block its access to specific tools. This control model aligns with the principle of least privilege, ensuring that each agent receives only the rights strictly necessary to fulfill its tasks, thereby reducing the risk surface within corporate environments.

Supported agents can be onboarded as team members on the same level as human users, leveraging the internal Multiparty Computation (MCP) and Agent‑to‑Agent (A2A) protocols that already operate under the hood of AMP. Additionally, the platform allows the capture of external agent sessions—for example, when an OpenAI model is invoked from an outside application—and associates those interactions with Jira issues. In this way, every step taken by an agent is logged in a ticket, facilitating auditability, troubleshooting, and accountability while providing a complete trace of AI contributions to business processes.

Rovo Work Extension and Long‑Task Planning

Within the AMP ecosystem, the Rovo Work feature introduces the ability to create long‑duration tasks with a revisable plan, automatic notifications, and limited permissions. These tasks can be assigned to agents or mixed teams, and progress is updated in real time within the project console. The combination of structured planning and granular access control enables complex projects involving AI to maintain clear traceability without sacrificing the operational flexibility needed for rapid innovation.

The primary challenge highlighted by Atlassian is that the visibility and governance promised by AMP depend on effective integration with external providers. While the architecture is built to log actions and enforce policies, the quality of those logs and the ability to apply real‑time controls are contingent on each third‑party agent exposing the necessary data through AMP connectors. In practice, administrators now have a unified console where they can audit, suspend, and reconfigure agents, but the full effectiveness of the governance model will require AI partners to adopt the integration mechanisms defined by Atlassian.