Scaled to 5M+ custom AI agents in one month
Significantly reduced customer costs for agent loops
Improved latency across workflows
Boosted quality aided by many agent use cases
Defaulted custom agent creation to Gemini Flash
Atlassian is scaling its multi-model artificial intelligence capabilities to millions of users by using Google Cloud and Claude to orchestrate powerful agentic workflows.
Atlassian is scaling its multi-model artificial intelligence capabilities to millions of users by using Google Cloud and Claude to orchestrate powerful agentic workflows.
Embracing openness: Atlassian’s mission to scale trustworthy AI
Atlassian’s mission is to unleash the potential of every team with world-class collaboration and productivity tools. Serving over 350,000 customers worldwide—powering everything from everyday business operations to rockets heading into space—the company acts as a vital bridge connecting technology and business teams through tools like Jira and Confluence.
Over the last two years, the rapid adoption of AI has presented a unique mix of customer demands and internal transformations. In its early days of using generative AI, Atlassian relied on a single-vendor model architecture. However, true to its 20-year history as a provider of open solutions built on integration, Atlassian quickly realized that a one-size-fits-all approach wouldn't address the needs of its diverse customers.
To meet customers where they are, Atlassian needed to build a flexible, model-agnostic platform that could scale efficiently while adhering to specific compliance, governance, and regulatory requirements.
The commitment to deploying AI to hundreds of thousands of global users immediately surfaced challenges around scale, complexity, and trust. Atlassian needed to maintain low latency, maximize cost efficiency, and empower its internal product developers to adapt rapidly in an environment where long-term planning wasn’t an option.
Powering the right tools for each unique job
To address the challenges, Atlassian built an internal AI model gateway on Gemini Enterprise as the primary routing path, providing the infrastructure to seamlessly manage and interconnect multiple AI workloads. Atlassian scales its AI framework using Google Kubernetes Engine (GKE) alongside Google Cloud’s high-performance GPUs and TPUs. This infrastructure provides a foundation for Atlassian's unique toolbox that helps users choose the best model for their specific job.
When an AI roadmap lasts no more than three months, flexibility is your greatest competitive edge. Running our AI model gateway on Google Cloud gives our development teams the ability to experiment, iterate, and scale new capabilities at lightning speed.
Sherif Mansour
Head of AI, Atlassian
Atlassian's Agent Platform is powered by multiple foundation models using Google Cloud AI infrastructure, assigning specific models to specific tasks. Claude is used in complex, long-running developer or agentic use cases. It serves as the primary engine behind Rovo Chat (Atlassian's in-app contextual chat), executing massive tool and skill calling across thousands of internal developer workflows, as well as powering internal developer tools like Rovo Dev (now Rovo CLI) and Rovo Max mode for complex multi-step tasks. Atlassian engineers and non-technical teams alike use Claude Code to rapidly prototype and commit code.
"In the world of AI, you cannot use one tool for everything," says Sherif Mansour, Head of AI at Atlassian. "Working with Google Cloud and Anthropic gives us the ultimate enterprise toolbox to automatically route the right workload to the right model at the perfect time to maximize efficiency and performance."
Gemini Flash models serve as the out-of-the-box default model for Atlassian's custom agent platform within Rovo Studio. Gemini Flash provides the broad reach and speed required to handle unlimited general-purpose agent workflows, such as low-latency question-and-answering across billions of enterprise documents.
"When an AI roadmap lasts no more than three months, flexibility is your greatest competitive edge," says Mansour. "Running our AI model gateway on Google Cloud gives our development teams the ability to experiment, iterate, and scale new capabilities at lightning speed."
"This multi-model ecosystem allows models to play off of each other and improve the overall performance of Atlassian's agents." By setting up an LLM-as-a-judge framework, different model architectures review internal workflows to eliminate inherent biases and benchmark performance.
"The diversity of models critiquing each other is excellent for customers," adds Mansour. "With more people thinking about a given problem from different perspectives, we get better outcomes."
Five million executed agents and a future built on context
The collaboration between Google Cloud and Anthropic has benefited Atlassian and its customers in many ways.
For example, migrating the custom agent platform to run natively on Gemini Flash models helped Atlassian achieve an exceptional balance of performance and affordability. The company has successfully scaled its ecosystem to execute more than five million custom AI agents within business workflows in just one month using a combination of Claude and Gemini.
With Google Cloud, Atlassian decreased the operational costs of running continuous agent loops—efficiencies that it then passed on directly to customers. Quality improved across a substantial number of use cases, while latency remained completely on par or improved depending on the use case.
We have significantly reduced operational costs for our customers running continuous AI workflows on our platform. By using Gemini and Claude through Google Cloud, we hit a rare trifecta: costs dropped, quality improved, and latency remained optimized.
Sherif Mansour
Head of AI, Atlassian
Finally, the model gateway allows Atlassian to easily upgrade models or swap workflows when vendors deprecate older versions, keeping the organization completely at the forefront of AI enablement.
"We have significantly reduced operational costs for our customers running continuous AI workflows on our platform," says Mansour. "By using Gemini and Claude through Google Cloud, we hit a rare trifecta: costs dropped, quality improved, and latency remained optimized."
Looking ahead, Atlassian is focusing its AI strategy on workflows and context. Rather than looking purely at individual token-maxing productivity, Atlassian is optimizing shared team workflows where automated agents can triage bug reports, review legal contracts, or cross-reference data seamlessly.
Central to this future is Atlassian's Teamwork Graph, an engine containing billions of objects that securely connects Atlassian apps with third-party environments like Google Workspace, Figma, and Salesforce. This keeps user data strictly permissioned while giving AI models the deep, real-time context required to answer complex queries in seconds.
Atlassian is a leading provider of team collaboration and productivity software, delivering widely recognized tools like Jira and Confluence to bridge technology and business departments. The company's foundational mission is to unleash the potential of every team, serving over 350,000 customers globally ranging from everyday operations to space exploration initiatives.
Industry: Technology
Location: Australia
Products: Gemini Enterprise, Gemini, Google Kubernetes Engine, Google Cloud GPUs, Google Cloud TPUs, Google Workspace, Claude on Google Cloud
About Google Cloud partner — Anthropic
Anthropic is a frontier AI company whose mission is to steer the trajectory of AI to advance human progress. We are best known for building Claude, the intelligence platform trusted by millions of people and businesses worldwide. Anthropic is a public benefit corporation—a for-profit committed to operating in service of social and public good—and controlled by a Long Term Benefit Trust, a group of independent experts in AI safety, national security, public policy, and social enterprise.
