OpenAI / 6 October 2026

Atlassian Deepens OpenAI Partnership Across Enterprise Work

Atlassian and OpenAI expanded their partnership on 6 October, combining model access, internal adoption and plans to connect AI agents more closely to enterprise work. OpenAI said more than 3,000 Atlassian developers use Codex, while Atlassian’s Rovo products draw on OpenAI models and the company’s Teamwork Graph. The agreement also supports further integration between ChatGPT and Atlassian applications. This gives OpenAI both a substantial named internal user and a route into workflows that already hold projects, documents and organisational context. Neither company disclosed the contract’s value, and the developer count measures adoption rather than a verified improvement in delivery speed or a corresponding number of newly purchased subscriptions.

The two companies describe a relationship stretching back to 2023, with the expanded agreement covering newer model access and product collaboration. OpenAI itself uses Jira, providing a second direction of operational experience rather than a simple supplier-to-customer arrangement. Atlassian’s Don Baron set out a broader ambition in which work items, agent sessions and human reviews can remain connected. These are useful distinctions for customers evaluating what they can deploy immediately. Access to an existing plugin, an internal development practice and an intention to make agents more autonomous represent different stages of implementation, with different requirements for administration, purchasing and assurance of the resulting work.

The integration programme includes Atlassian services such as Jira and Confluence, with deeper assignment and tracking of work discussed as future development. Atlassian’s account emphasises organisational context, while OpenAI’s Nikunj Handa describes helping users “move from understanding their work to taking action.” The economic mechanism is that a model can become more useful when it knows which project, document and approval process a request belongs to. That does not eliminate the need to respect the permissions of the underlying application. It also means that model quality alone cannot determine the customer outcome: incomplete records, conflicting instructions or poorly defined approvals can still derail an otherwise capable agent.

Atlassian also points to DX as a way to understand software delivery and the effects of AI adoption. Measuring those effects requires more than counting seats or generated code. A team could produce changes faster while increasing review work, operational failures or maintenance costs; alternatively, the principal benefit could be reducing interruptions rather than increasing output. These are measurement considerations, not reported outcomes from the partnership. The announcements do not disclose a controlled before-and-after productivity study for the 3,000 developers. Customers therefore have evidence of meaningful internal use and a stated measurement approach, but not a transferable percentage saving they can safely apply to their own engineering budgets.

The arrangement also raises a question about where enterprise value accumulates. OpenAI supplies model capability and a general working interface, while Atlassian retains the applications and records around which many teams organise their work. Each side becomes more valuable when the other reduces the effort required to complete a task, but each also has an interest in remaining the place where customers initiate and supervise it. This is a strategic inference from the product structure, not evidence of a commercial dispute. The disclosed partnership gives both companies a way to improve integration while leaving contract economics, revenue sharing and the long-term allocation of customer ownership unspecified.