How Atlassian builds AI agents teams can trust with Claude and Google Cloud

Overview

Atlassian, whose products including Jira and Confluence serve over 350,000 customers, built an internal AI model gateway on Google Cloud that routes complex, long-running agentic work to Claude while simpler high-volume tasks go to Gemini Flash. Claude powers Rovo Chat (Atlassian's in-app assistant used by millions), Rovo CLI (its primary developer AI experience), and Rovo Max, an early-access mode that plans and executes complex multi-step work across Atlassian's Teamwork Graph. More than 5 million AI agents now run in customer business workflows each month.

The challenge

Atlassian's customers want to hand consequential, repeatable work to trusted agents - reviewing incoming contracts, triaging support tickets, surfacing sales leads - rather than just chatting with an assistant that still leaves the work to a person. That kind of work requires an agent to use Atlassian's own product capabilities from inside the product the way a person would, and trust is hardest to earn on complex, long-running tasks where a wrong answer compounds the longer an agent works.

The solution

Atlassian built an internal AI model gateway centered on Google's Gemini Enterprise Agent Platform as the primary routing path, scaled on Google Kubernetes Engine with Google Cloud's GPUs and TPUs, that routes workloads to the best-performing model. Claude handles the most complex and long-running work: it is one of the models behind Rovo Chat (the in-app assistant used by millions), powers Rovo CLI (Atlassian's primary developer AI experience), and is behind Rovo Max, an early-access mode that takes a goal, draws on Teamwork Graph (Atlassian's connected map of a customer's app objects), and plans, reasons and executes across Atlassian tools inside a managed environment where admins control package and network access. Gemini Flash is the default for general-purpose, high-volume customer agents, while Claude carries the complex, long-running work; teams write their own evaluations and benchmark models on quality and cost before deploying.

Reported business value

More than 5 million AI agents now run in customer business workflows each month, a figure that keeps climbing. Rovo Chat serves millions of users with Claude as a base agent model. Moving high-volume custom agents to Gemini Flash while reserving Claude for complex work reduced operational costs, improved quality, and kept latency optimized. In an internal demo, a Rovo Max agent turned a Jira board into a finished Instagram-style video reel by pulling in linked Figma designs and Google Docs via Teamwork Graph, writing its own script, and stopping only where it needed a human-provided Instagram account.