Atlassian's Agents Write 50% More in Jira. Count What the Humans Closed.

Atlassian's shareholder letter is the week's number, and it is the second time in one week that a vendor put the agent on the human's seat. This is a Ship Story: the decision, the result as reported, and the ratio the letter leaves out.

The short version

Mike Cannon-Brookes's Q1 FY27 letter of November 5: subscription ARR up 22.5%, cloud revenue up 30.5%, RPO up 44.4% to $4.7 billion, users of Rovo Agents 2x in six months, more than 2 million monthly active users on the Atlassian MCP Server (8x since March), more than 15 million MCP calls a day (15x since March), users creating 50% more in Jira and 75% more in Confluence after adopting the MCP Server, and MCP adopters growing ARR at about 2x the rate of non-adopters. The decision was to open Jira and Confluence to any coding agent and let it write; the letter's own summary is that agents are increasing engagement rather than replacing the platform, and customers continue to add seats. The agent is a free writer on a paid human seat, and Salesforce priced the same arrangement at $50 per user on November 4. It holds as long as the human closes what the agent opens, and the letter counts what was created and never that pair. On the other side of the ledger, 13 of my 39 scheduled agents produced output nothing read and 30 of 51 commits in my agent-maintained library have no attributable author. The number to look for next quarter is closed per created, and the seat line on the day the agent is the reader too.

What the letter says

The numbers come first in the letter and I will keep them in Atlassian's order. Subscription ARR grew 22.5% year over year. Cloud revenue grew 30.5%. RPO grew 44.4%, to $4.7 billion, with current RPO up 28%. Users of Rovo Agents doubled in six months, and Rovo assisted actions per user are up more than 3x since the start of the year.

More than 3,000 customers are on Teamwork Collection, 7x the prior year. Those customers complete 35% more work items per Jira user than customers on Jira alone and run twice as many agents. Service Collection passed 70,000 customers with ARR growing north of 30%.

Then the agent section, which is the one I care about. The Teamwork Graph maps more than 250 billion objects and relationships, up from 150 billion six months earlier. Any agent or coding assistant can tap it through the Atlassian MCP Server. More than 2 million monthly active users engage with that server, up 8x since March, nearly half from enterprise customers. More than 15 million MCP calls a day, up 15x since March, with Codex, Claude Code, and Cursor creating Jira work items, updating statuses, and moving work forward. After adopting the MCP Server, users create 50% more in Jira and 75% more in Confluence on average. 98% of MCP users are active on the platform in the same month. Customers using the MCP Server grow ARR at about 2x the rate of non-adopters.

And the two sentences that sit next to each other and make the story. "Agents aren't replacing our platform. They're increasing engagement and amplifying its value." A few paragraphs earlier: "Customers continue to add seats in Jira and Confluence."

One more line for the people who read this site for the forward deployed series: Atlassian expanded its Forward Deployed Engineering program this quarter, with Expedia and Reddit named as customers already seeing value.

The Ship Story

The decision: open the system of record to any agent and let it write. Not a Rovo-only door, an MCP server any coding agent can call, with the writes counted as platform engagement.

The result, as reported: 15 million calls a day, half again as many Jira items per user, longer and larger commitments, and a 44% RPO line.

Read it as a pricing decision and it is a clean one. The seat is still the unit Atlassian sells. The agent does not buy a seat. The agent writes into the human's, and the human's seat becomes worth more, because the human's Jira now has 50% more in it. The SaaS to AI Business Models thesis is that the margin structure follows the pricing model, and this is the per-seat model finding a way to charge for agent work without pricing it: the agent is a free writer on a paid seat.

It is the second time this week. Salesforce Put a $50 Seat Next to the Credit on Agentforce Coworker. The Seat Is the Budget's Unit. covers the November 4 release that put a per-user, per-month price on an agent product. Salesforce priced the seat. Atlassian counted the writes. Two vendors, one week, both putting the agent on the human's seat from opposite sides.

The number the letter does not give

The arrangement holds as long as the human closes what the agent opens.

The letter counts creation. Items created per user after MCP adoption, up 50%. It counts completion once, for Teamwork Collection customers, 35% more work items completed per Jira user. What it never counts is the pair: items an agent created against items a human closed, read, or reversed. Every figure in the agent section is a volume figure, and volume is what an agent produces for free.

I have the other side of that ledger, and it is not flattering. 70% of LangChain's Agent Runs Are Scheduled. Count the Ones Somebody Read. has the audit: of my 39 appointed agents, 13 were orphans producing clean output that nothing outside the fleet read. 43 Skills, 39 Auto-Activate. Kill the Skill Count. Keep the Four You Type. has the library: 30 of 51 commits since March 30 with no author I can attribute. An agent-created Jira item that no person opens is the same orphan run, with a ticket number and a seat underneath it.

The 2x line deserves the same reading. Customers using the MCP Server grow ARR at about twice the rate of non-adopters. The letter does not say which came first, the rich graph or the growth, and its own flywheel paragraph describes customers who were already expanding across apps and adopting Collections. A customer growing fast connects more tools. That it then shows up as an MCP adopter growing fast is not evidence that the MCP Server did it.

Closed per created

The seat gets fatter while the human is the reader. That is the whole result, and it is a real one; a 44% RPO line is not engagement theater.

The question is what happens to the seat on the day the agent is the reader too, when the item Codex created is triaged by Rovo, assigned by an automation, and closed by another agent, and the human's seat has nothing in it that a human touched. I ran that migration from the other direction. Field Report: What Broke When We Killed Our Per-Seat Tier has what the seat line did when the work was priced instead of the person: 14 of 35 mid-market accounts churned, and by month 22 outcome revenue was 91% and blended gross margin was back to 71%. The seat is what you sell while the human is still the reader.

Two numbers to look for in next quarter's letter. Closed per created: of the work items agents create, the share a person closed, read, or reversed within the quarter. And seats per MCP-active customer, quarter over quarter, which is the first line that moves when the agent starts reading.

This week, if you run a platform with agents writing into it: pull the items your agents created in October and count how many a human has opened. If the share surprises you, it will surprise your customers first.

Related answer: Does agent-created activity in Jira count as adoption?

Sources: Mike Cannon-Brookes, "Our Q1 FY27 letter to shareholders," Atlassian, November 5, 2026, read in full; every ARR, revenue, RPO, Rovo, MCP, Teamwork Graph, and Collection figure, and both quoted sentences, are Atlassian's as the letter reports them; the two customer quotations in the letter were not used. Falk Gottlob, "Salesforce Put a $50 Seat Next to the Credit on Agentforce Coworker. The Seat Is the Budget's Unit." (the $50 price, as Salesforce's November 4, 2026 release reports it), "70% of LangChain's Agent Runs Are Scheduled. Count the Ones Somebody Read." (the 13 of 39), "43 Skills, 39 Auto-Activate. Kill the Skill Count. Keep the Four You Type." (the 30 of 51), and "Field Report: What Broke When We Killed Our Per-Seat Tier" (the 14 of 35, the 91%, and the 71%).