Atlassian’s New “Agentic Multiplayer Protocol” Wants to Answer One Question: Who Actually Wrote This Code?
October 7, 2026
Atlassian has launched the Agentic Multiplayer Protocol (AMP), a new foundation for how humans and AI agents work together across its platform — with agents getting their own identity, scoped authority, and a version history you can audit. Why it matters: most enterprises still can’t tell what an AI agent wrote versus a human, and that blind spot is what’s holding back serious AI adoption in software teams.
What Atlassian announced
AMP is Atlassian’s framework for agents “taking part in multiplayer work” — meaning agents participate in projects with an identity, limited permissions, shared context, assigned tasks, and results a human can review. It shipped alongside three related pieces:
- Teamwork Graph — indexes source code down to functions, symbols, and classes, so developers can search across Bitbucket and GitHub without cloning a repo. Crucially, it provides version history showing exactly what a human wrote versus what an agent did, across Claude, Codex, Figma, and Rovo.
- Rovo Work — a new mode in Rovo Chat that handles complex, multi-step tasks with human oversight. It executes across Jira, Confluence, and connected tools, and can run for hours in a secure sandbox toward a goal the user approved.
- EU AI Inference — restricts LLM processing to models hosted within the EU for data-residency needs.
The company also promised non-human identity (NHI) controls — limits on what AI can see and what access agent accounts get — rolling out “soon.”
Why attribution is such a hard problem
Git records an author and a committer, but when a developer runs an agent locally, the commit names the developer — the agent vanishes into the record. Developers edit agent output, agents run side by side, and provenance that lived in commit text gets lost when history is rewritten. As Mike Wilkes, enterprise CISO at Aikido Security, put it: Git can tell you who committed a change, but that’s increasingly different from who — or what — actually wrote the code.
Why it matters: the “everything AI, tested” take
The honest version of this story: the labeling is table stakes, but the measurement layer underneath is the real prize. Reliable attribution lets a CIO finally measure agentic development instead of counting AI licenses — comparing human versus agent-assisted changes on cycle time, defects, rollbacks, and cost per accepted change. IDC’s Adam Resnick notes the value goes beyond who typed a line: enterprises want visibility into where judgment happened and whether meaningful human review took place, not just that someone clicked approve. And one caveat deserves scrutiny: Atlassian says it traces every change to the agent and “the person who set it up” — but the person who set up an agent isn’t always the person who approved its work. The approver owns it; version history should make that visible, not blur it.
Frequently asked questions
What is Atlassian’s Agentic Multiplayer Protocol?
It’s a framework that gives AI agents an identity, scoped permissions, shared context, and reviewable results across Atlassian’s platform — so agent work is tracked as agent work, not flattened into a human’s commit history.
Is AMP available now?
Atlassian says the AMP capabilities, Teamwork Graph, Rovo Work, and EU AI Inference are available immediately; the non-human identity controls are promised “soon.”
Which coding tools does the attribution cover?
Teamwork Graph’s version history covers Claude, Codex, Figma, and Rovo, across both Bitbucket and GitHub.
Doesn’t Git already track who wrote code?
Only nominally. Git tracks the committer, which is almost always the human developer — even when an agent did the work. Attribution usually lives in editable commit text or disappears entirely.
Why should a business care who wrote a line of code?
For risk-based review and for economics. Knowing what agents produced lets companies measure whether their AI investment is actually moving cycle time, defect rates, and delivery — rather than guessing.
Sources: InfoWorld