AI just went multiplayer. Your engineering metrics have to count every player.
Atlassian wants AI agents to stop working in private chat windows and start working as visible, accountable members of the team. That is the right move. It also raises the question every engineering leader will be asked next: now that we can see what agents did, was it worth it?
At Team ’26 Europe in Amsterdam on October 7, Atlassian announced the Agentic Multiplayer Protocol (AMP). The idea is simple to state and hard to build: AI agents should take part in shared work the way people do, inside the same governed system, rather than as tools each person consults alone.
AMP is grounded in Atlassian’s Teamwork Graph, which the company says now links more than 250 billion objects and relationships across work items, documents, people and code. Around it, Atlassian shipped a rebuilt MCP server with more than 200 tools (already handling around 15 million tool calls a day), Rovo Work for multi-step tasks that humans review and approve, and code indexing down to functions, symbols and classes across Bitbucket and GitHub, with GitLab support announced as coming.
Strip away the product names and AMP defines five things every agent needs to be a real participant in team work.
A “diplomatic protocol”
That is how Atlassian CEO Mike Cannon-Brookes framed AMP: less a technical standard, more a set of rules for how humans and agents treat each other at work.
Engineering is where single-player AI broke down first. A developer accepts a Copilot suggestion, pastes something from ChatGPT, hands a task to Claude Code, and lets an autonomous agent open a pull request, all in the same afternoon. Most of that happens in private sessions. When the code lands, nobody can say with confidence which parts came from a person, which came from an agent, and what context either of them had.
Multiply that across a team running several coding agents from different vendors in parallel, and you get agent sprawl: lots of output, no shared record of who or what did the work, on whose authority. Atlassian’s own leadership has said this lack of visibility into code origin is holding enterprises back from adopting AI further.
The gapAMP is mostly about the first two rungs of a ladder every engineering organization has to climb. The third rung is the one the CFO and the board care about.
Analysts covering the launch made the same point from a different angle. Coverage in InfoWorld noted that attribution often lives in commit messages, which developers can edit, disable or lose when history is rewritten, and that most real code is a blend of human and agent work. Attribution is easy at the edges and hard in the middle, which is exactly where most of the work now happens.
Reliable attribution lets CIOs “finally measure agentic development rather than merely count AI licenses and token consumption.”
Mike Wilkes, enterprise CISO at Aikido Security, quoted in InfoWorld
That is the real prize. Counting seats and tokens tells you what you spent. Measuring agentic development tells you what you got.
What to measureOnce agents have identities and owners, you can finally measure them like teammates. These are the signals we recommend engineering leaders build around.
AMP is strongest where Atlassian already lives: Jira, Confluence, Bitbucket, and the Teamwork Graph behind them. That is a big footprint. But very few engineering organizations run on one vendor. Code sits in GitHub, GitLab and Azure DevOps. Planning lives in Jira, Linear or both. Agents come from Anthropic, OpenAI, GitHub, Cursor and Atlassian, often in the same team.
That is why the measurement layer has to be neutral. Identity and governance can live inside each platform. The question of whether your AI investment is paying off has to be answered across all of them, with the same definitions, for every team.
Giving agents a name tag is how you see them. Measuring their outcomes is how you decide whether to keep them.
Alex Circei, CEO, Waydev
AMP is a meaningful step. When one of the largest collaboration platforms in software says agents need identities, owners, permissions and audit trails, the debate about whether agents are teammates is over. They are.
The harder conversation starts now. Once you can see every player, you will be asked how each one is performing, what they cost, and whether the team as a whole ships better software than it did a year ago. Engineering leaders who can answer that with data will set the pace for everyone else.
Measure your whole human-agent team
See how engineers and agents across every tool are changing your delivery, quality and cost, in one view built for the multiplayer era.
Book a demoSources: Atlassian, Agentic Multiplayer Protocol; InfoWorld; SiliconANGLE; diginomica; NAND Research. Coverage of Team ’26 Europe, October 2026.
Ready to unlock your SDLC productivity?
