Atlassian Launches Agentic Multiplayer Protocol to Enable Humans and AI Agents to Work Side by Side

Atlassian Introduces Framework for Human and AI Agent Collaboration

Atlassian unveiled its Agentic Multiplayer Protocol at Team '26 Europe in Amsterdam, marking a shift in how enterprises approach human and AI agent collaboration. CEO Mike Cannon-Brookes emphasized that "working with AI isn't single player any more—it's multiplayer," positioning AMP as both a technical and behavioral framework2

2

. The protocol defines how AI agents participate in work with assigned identity, scoped authority, shared context, and reviewable results across the Atlassian platform1

1

.

Jamil Valliani, head of AI products at Atlassian, told reporters that lack of visibility into code origin "is holding enterprises back from further adopting AI"1

1

. The agentic AI age requires businesses to understand context as the fuel that powers AI models, with Cannon-Brookes noting that "no matter how capable a model gets, if it doesn't reliably know why you made a decision back in 2022, it can't help you in 2026"2

2

.

Non-Human Identity Controls and Governance Framework

Atlassian introduced a new identity class for non-human participants, including AI agents, apps, and service accounts. Admins gain full control to extend or revoke permissions for these non-human identity controls, with complete audit trails tracking activity and token spend4

4

. The system discriminates between four interaction patterns: human activity, agent activity with non-human identity, agent on behalf of human, and human assisted by agent4

4

.

The Teamwork Graph indexes source code down to functions, symbols, and classes, providing full attribution and accountability with version history showing exactly what humans wrote versus what agents did across Claude, Codex, Figma, and Rovo1

1

. This AI code visibility addresses enterprise concerns about differentiating AI-generated code from human contributions, though analysts note challenges remain. Adam Resnick, research manager at IDC, emphasized that "the real value goes beyond knowing who typed a given line" and requires understanding where humans exercised judgment on consequential decisions1

1

.

Expanded OpenAI Partnership with Multi-Model Approach

Atlassian deepened its OpenAI partnership, integrating GPT-6 Astra and GPT-5.6 series models into its platform while maintaining a multi-model approach3

3

. The arrangement includes a spend commitment to OpenAI models and technology, but GPT-6 Astra is not becoming Rovo's default model3

3

. Rovo runs an internal AI gateway that routes dynamically across OpenAI and other providers, balancing capability, speed, and cost per task.

The rebuilt Atlassian Model Context Protocol server now handles more than 15 million calls daily, exposing 220-plus tools across Jira, Confluence, Bitbucket, Loom, and Goals3

3

. Internal testing showed up to 25% fewer tokens on comparable Jira and Confluence work. The MCP server is Atlassian-hosted, secured with OAuth 2.1 and PKCE, and enforces each user's existing permissions at query time, ensuring agents cannot access data beyond user permissions3

3

.

Rovo Work Enables Complex Multi-Step Tasks

Atlassian introduced Rovo Work, a new mode in Rovo Chat that handles complex, multi-step tasks with human oversight to review and approve them1

1

. The system executes across Jira, Confluence, and connected tools, running for hours in a secure sandbox to achieve user-approved goals. Valliani explained that when users send a query to Rovo Work, they're "giving Rovo permission to go and actually unleash itself fully"5

5

.

Rovo can go beyond straightforward task completion by training itself on unfamiliar tasks. In one example, when asked to create an Instagram-ready reel, Rovo researched the right instructions, learned the required output format, and obtained tooling for video synthesis5

5

. This represents a shift from the standard AI approach of checking existing training data to actively learning new capabilities. For power users, Rovo can generate custom skills as part of its work, though this remains a separately supported capability5

5

.

Loom Integration Simplifies AI Collaboration Tools

Atlassian introduced Loom Record for Agent, replacing traditional prompt engineering with visual instruction. Users can record videos showing what they want, drawing circles around UI elements, clicking buttons, and entering information into fields while providing verbal instruction5

5

. Loom captures both the visual references and spoken instructions, formatting them into prompts that give agents the same context two humans would have looking at a screen together5

5

.

The platform also includes EU AI Inference, which restricts LLM processing exclusively to models hosted within the EU1

1

. Data governance remains central to the multiplayer system, with Data Security Policies and Data Loss Prevention enforced through Atlassian Guard across MCP connections. Admins receive domain allowlists, IP allowlists, per-scope read/write/search controls, and org-wide audit logging from the Admin Console, including the ability to disable writes entirely3

3

.

Measuring Enterprise AI Adoption Value

Mike Wilkes, enterprise CISO at Aikido Security, noted that reliable attribution lets CIOs "finally measure agentic development rather than merely count AI licenses and token consumption"1

1

. Organizations can compare human versus agent-assisted changes on cycle time, review effort, rework, defects, and rollbacks to calculate cost per accepted change. This visibility proves critical for CIOs making funding and AI-assisted coding strategy decisions in the agentic AI age.

Cannon-Brookes emphasized that managing the pace of progress is critical, noting that "when every person, every agent starts to move faster and faster, new challenges arise" because it's not just one person moving faster but a multiplication of teams increasing outputs2

2

. The Atlassian platform aims to serve as a company's "central nervous system" by providing context, collaboration, and confidence so workers can quickly access information when needed2

2

.