The teamwork graph is now the largest claim in enterprise search. Atlassian told shareholders on 5 November 2026 that its Teamwork Graph maps more than 250 billion objects and relationships, up from 150 billion six months earlier, and that humans and agents collaborate on its platform more than 10 million times a month, a figure that doubled in six months. A graph at that scale is a record of the company. The open question is who keeps that record true when a fact changes.

The teamwork graph numbers are usage numbers

Atlassian published the figures in its Q1 FY27 letter to shareholders. Rovo agentic automations are up 9x year over year. Users of Rovo Agents have grown 2x in the last six months. Rovo assisted actions per user are up more than 3x since the beginning of the year. More than 2 million monthly users engage with the Atlassian MCP Server, up 8x since March, with more than 15 million MCP calls a day, up 15x since March. More than 80 percent of highly regulated cloud customers have Rovo enabled. These are company reported usage figures from a shareholder letter, not independent measurements of answer quality, and the post treats them as such.

The Graph spans seven types of context, in Atlassian's list: knowledge, work, people, communications, code, assets, and now data. It connects to more than 80 third party tools out of the box, and more than 60 percent of customers connect four or more types of context. Atlassian calls the Graph a living memory of each customer's organization. Kaplan's Executive Director of Enterprise Agility and Delivery, Jose Ramirez, is quoted in the letter saying that a good model powered by Atlassian's organisational context will always outperform the best model in the world working in a vacuum.

A rich graph predicts adoption. It does not prove truth.

Atlassian reports that customers with a rich graph are twice as likely to become deep Rovo adopters, and that those adopters grow annual recurring revenue more than 2x faster than customers with light usage. That is a correlation between graph depth and Atlassian revenue growth. It does not show that the relationships in the graph are current, that a superseded decision has left the graph, or that two conflicting definitions of the same term were resolved by a named person. A relationship can be wrong at scale as efficiently as a document can. When an agent acts on a wrong edge, it does not produce a wrong paragraph. It routes work, assigns a person, or updates a status, and the error enters the work itself.

The expansion is also reaching colder content. On 11 November 2026 SmiKar announced Forage for Squirrel, which brings live SharePoint documents and documents archived to Azure Blob Storage into one searchable environment and extends Microsoft Copilot to archived content. Forage verifies SharePoint permissions before making content available and records every search and retrieval against the authenticated user. Its AI classification recommendations do not automatically change document classifications; administrators review the suggestions and decide which rules to enable. SmiKar Sales Director Mark Smith is quoted in the release saying that archiving should not mean losing visibility of information. The direction is sound. It also means a contract archived several years earlier, SmiKar's own example, can now answer a question beside the current agreement, with nothing in retrieval alone to say which one governs.

Every collaboration now carries a meter

Scale has a price attached. An alternatives comparison checked against Atlassian's published Rovo credit rate card on 27 October 2026 reports that a quick Rovo Chat answer costs 10 credits and that extra usage billing for Rovo credits takes effect on 3 December 2026 at $0.01 per credit. The comparison is a vendor page, not an Atlassian source, and the rate card is the settling source. The point stands on the primary figures alone: at 10 million collaborations a month and 15 million MCP calls a day, each retrieval is a billed, logged event. A buyer can count the collaborations. Counting is not the same as auditing what each collaboration relied on.

The Graph Receipt Test

A teamwork graph earns trust the way a record does, with a receipt for each edge. First, an edge receipt: every relationship an agent relies on carries its source, its date, and the version it superseded, not only the documents at either end. Second, a named approver: when two sources define the same term or the same decision differently, a named person resolves the conflict and the superseded version stays visible. Third, correction propagation: when a fact is corrected, the correction reaches derived summaries, standing agent instructions, automations and queued MCP actions that were built on the old version. Fourth, an archived edge check: content that enters search from an archive carries its age and its status, so an old agreement is read as history until a person promotes it as current. This is the same standard Jilu applies in its own work on the compiled record and on governed ontology.

Atlassian has shown that a company will connect its tools when the graph makes agents more useful, and the adoption figures in the shareholder letter are real evidence of that demand. Usefulness is not the final test. A graph that no one can correct is a faster way to repeat an old decision. The companies that benefit most from a teamwork graph will be the ones that treat each edge as a claim with a source, a date and an owner, and that keep a person in charge of changing it. How Jilu builds that record is set out in How Jilu Works.