Atlassian AMP Targets AI Code Attribution Across Enterprise Workflows

InfoWorld reported that Atlassian has launched Agentic Multiplayer Protocol, or AMP, as a platform layer meant to show enterprise teams where software work comes from when human developers and AI agents are both changing code.

The release addresses a practical visibility problem rather than a simple human-versus-machine split.

Large codebases now contain work shaped by developers, copilots and task agents, and Atlassian is positioning AMP as the record that lets IT, security and engineering leaders see each participant's role before they approve broader AI use.

Jamil Valliani, Atlassian's head of AI products, told InfoWorld that weak origin tracking is one of the obstacles slowing enterprise adoption.

"For most enterprises, this is what is holding them back from further adopting AI," he said.

AMP is designed to give agents an identifiable place inside Atlassian workflows.

The company describes the protocol as a foundation for collaboration between people and agents, with identity, bounded authority, shared context, assigned work and reviewable outcomes attached to the agent's activity.

That makes the system less about a standalone coding assistant and more about a controlled collaboration layer across the tools enterprises already use to plan, write and manage software.

Teamwork Graph is the core technical mechanism in the announcement.

It indexes code at the level of functions, symbols and classes, so developers can search across Bitbucket and GitHub without first cloning a repository.

For engineering managers, the more consequential promise is traceability: version history is supposed to separate work written by a person from work performed by an agent across Claude, Codex, Figma and Rovo.

That attribution layer matters because mixed code origin changes review and accountability.

A security lead may need to know whether an agent touched a sensitive function, while an engineering lead may need to understand whether a design change moved through Figma, a repository and Jira as one coordinated task.

AMP gives Atlassian a way to keep those steps connected to a named human or agent identity rather than leaving the audit trail scattered across separate tools.

Atlassian is also extending Rovo Chat with Rovo Work, a mode built for longer, multi-step assignments.

Rovo Work can operate across Jira, Confluence and connected applications, but the workflow keeps human review and approval in the path.

The source describes the mode as able to run for hours inside a secure sandbox while pursuing a goal that a user has approved.

The rollout includes a regional control for AI processing.

EU AI Inference is intended to keep large language model processing limited to models hosted in the European Union, a feature aimed at organisations that must control where AI workloads are handled.

Atlassian says the announced capabilities are available immediately.

A second set of controls is still on the roadmap.

The company plans non-human identity capabilities that would examine what information is visible to AI systems and restrict the access granted to agent accounts and other non-human identities.

Those controls are expected in the coming months, leaving the immediate AMP launch focused on visibility, attribution and supervised agent work rather than the full access-governance package.