Atlassian and OpenAI expand partnership for Rovo and workplace AI agents

Atlassian and OpenAI announced an expanded partnership on October 6, 2026, to bring OpenAI models into Atlassian’s Rovo product and wider workplace platform. For organisations using tools such as Jira and Confluence, the agreement promises more ways for AI assistants to draw on their projects and documents. The companies say some connections already exist, while deeper agent functions remain under development.

Under the agreement, OpenAI says its models will power agents across Atlassian’s platform and Rovo. Rovo uses Atlassian’s Teamwork Graph, which links people, projects, documents and decisions to provide context for AI responses. The expanded access includes GPT-6 Astra and the GPT-5.6 series, according to OpenAI. Neither company’s public account gives the agreement’s value or duration.

What Atlassian and OpenAI customers can use now

OpenAI says its models already support reasoning in Rovo and across Atlassian’s platform. The companies also describe connections that let ChatGPT and Codex reach relevant Atlassian work through plugins and the Model Context Protocol, or MCP. Atlassian says those connections reflect existing user permissions. That matters because an assistant’s answer can draw on a team’s own work items and documentation, rather than a general prompt alone.

As an illustration, OpenAI describes a product manager asking about launch readiness while Rovo draws on Jira tickets, Confluence documents and discussions to identify blockers, missed milestones and decisions. The example explains the intended use of connected information; it is not a measured customer result. Atlassian’s partnership post quotes Datasite executive JR Harrell describing context-aware actions at his company, but provides no independent measure of the effect.

A wider connector for several AI clients

In a separate October 6 announcement, Atlassian said a rebuilt version of its MCP server is generally available, with more than 220 tools. The company says it receives more than 15 million calls a day. Its account describes connections spanning Jira, Confluence, Bitbucket, Loom and other work context, while listing Jira Service Management support as coming soon. The call volume is Atlassian’s figure, rather than an independently verified count.

The MCP server is designed for several AI clients, including Claude, ChatGPT and Cursor. VentureBeat reports, citing Atlassian correspondence, that GPT-6 Astra will not become Rovo’s default model and that Rovo routes work among OpenAI and other providers. The expanded OpenAI deal therefore sits alongside Atlassian’s use of other models, rather than establishing exclusive use of OpenAI technology.

Atlassian says its hosted MCP server uses OAuth 2.1 and PKCE by default, applies each user’s existing permissions, and offers controls for reading, writing and destructive actions. It also describes administrator controls and audit logs. Those are the company’s descriptions of its safeguards, not the findings of an independent security assessment. In an internal benchmark using Claude models, Atlassian says the new server used up to 25% fewer tokens than its previous version on the same Jira and Confluence work; that result does not establish a general customer productivity gain.

What the expanded agreement leaves open

The companies say their collaboration began in 2023. VentureBeat reports that much of the integration discussed this week predates the new announcement. It also reports that an OpenAI representative, speaking on background, characterised the expanded access as effectively a spending commitment by Atlassian for OpenAI models and technology. The representative did not provide contractual specifics, according to VentureBeat, and the public announcements do not disclose a financial figure.

The next proposed step is closer integration with work tracked in Jira. OpenAI describes agents picking up work items, tracking progress, recording decisions and reviewing results as possibilities being explored. Atlassian likewise says autonomous agent functions and coordination among multiple agents are still being designed, with human checkpoints envisaged. The companies have not established delivery dates or demonstrated customer results for those future functions. For now, the confirmed change is broader model access under the partnership and the existing ways to connect AI tools with Atlassian work context.