Director of One
[ ← Journal ]

Atlassian gave the agent an owner and a limit. Check one status.

Covers 2026-10-05 to 2026-10-07 Director of One Editorial

A director’s status says who owns the work, what is blocked, which decision is waiting, and whether the write-up matches the tickets. On October 6, Atlassian said it is starting to show an agent’s owner, a limit on what that agent may do, and a result a person can review, inside the products the team already uses.

The change is called AMP, the Agentic Multiplayer Protocol. It is Atlassian’s rule for an agent that joins the tools the team already shares.

What the October 6 posts say

Two Atlassian posts carry the claim, and they use different timing.

Mike Cannon-Brookes wrote that the AMP rollout begins on October 6. Jamil Valliani, head of product for AI, wrote the same day that the protocol is live across the Atlassian platform. Check your own Atlassian site before you tell the department which of those you have.

Both posts describe the same parts. The agent has an identity. A person scopes what it may do. It uses shared context and shared tasks. It returns a result a person can review.

Cannon-Brookes is specific about the record that comes with those parts. He describes an audit of who triggered the agent, what data it touched, and built-in human checkpoints. He names two ways a limit can be set. An HR agent inherits the strict access of the person prompting it. A helpdesk agent runs on its own dedicated service account.

The status claim is the one that can change a Monday meeting. Valliani writes that, at the portfolio level, agentic updates in Focus turn cross-project progress and goal signals into real-time summaries, so a leader can see risks and blockers without manual reporting. Cannon-Brookes writes that Jira Agent Sessions connect local and cloud agent sessions to the work item, so the team can see that work beside human work without chasing manual status updates.

Valliani also describes Rovo Work, for a task with several steps. You give it a goal, it proposes a plan, you shape the plan, and then it runs across Jira, Confluence, and connected tools, under a person’s review and approval.

Atlassian reports that people and agents already collaborate over ten million times a month on its platform. That figure is company-reported.

What is still ahead

Valliani writes that a non-human identity inventory is still being added. The inventory is meant to give every agent, app, and service account an identity, and to let an admin revoke access from one place. Cannon-Brookes lists that inventory among what the company is introducing this week. He uses “soon” for a different piece: bringing sessions from a developer’s editor or a terminal into Jira. Count each control only after it shows on your own admin screen.

A third post that day covers the OpenAI partnership. OpenAI models already power reasoning inside Rovo, and teams can connect ChatGPT and Codex to Atlassian context through Atlassian MCP. The later step, which Atlassian says is still being designed, is an autonomous agent that picks up a work item, runs tests, and syncs a local session back to the board. Reasoning over work the team already tracks is part of what the company says is available now.

Four lines for one project

Pick one project that crossed teams in the last week. Write these lines before you open the tool. Then mark each one present, partial, or missing.

  1. Owner. The person accountable for the agent that touched the work. Write a person’s name.
  2. Limit. What the agent was allowed to see and to change, in one sentence. If you cannot write the sentence, the limit is not set.
  3. Trace. The last change, who triggered it, and the work item where a person can open the result.
  4. Open decision. The call that is still waiting, and the name of the person who has to make it.

When lines 1, 2, and 3 are all present, they can replace the status email. Line 4 stays with a person. When the owner or the limit is still empty, hold the summary. Until those two lines exist, the summary is a chat transcript.

Fill the lines in this shape, then replace every role with someone on your project:

  • Owner: the manager who turned the agent on.
  • Limit: it may label new bugs. It may not change priority or close a ticket.
  • Trace: that label change, the person who triggered it, and the work item that holds it.
  • Open decision: whether the date moves, assigned to the director who owns the date.

If your project will not fit that shape, the summary is ahead of the record. Keep the summary in draft and send the four lines.

The same four lines work on a board other than Jira or Focus. The protocol belongs to Atlassian. The test belongs to the department.

What to do this week

  • Write the four lines for one live project on paper, before you open the tool.
  • Mark each line present, partial, or missing against what Jira, Confluence, or Focus shows.
  • When the owner or the limit is missing, keep that agent off any update that leaves the department.
  • Drop the status meeting for the lines the board already answers.
  • Leave line 4 as a named decision. That note is the meeting, if you still need one.

A consult block is five hours at $400 an hour with coaches on which meetings exist only to move status, who owns the outcome, and a delegation matrix for AI and people.

Sources

  1. The End of Single-Player AI · Atlassian · 2026-10-06
  2. The Atlassian platform is for multiplayer human-agent work · Atlassian · 2026-10-06
  3. Grounding Frontier Intelligence in Enterprise Context · Atlassian · 2026-10-06

Researched and drafted with AI assistance, checked against the sources above.

See how this could work in your day.

See a worked example →