Atlassian launches Code Context to boost AI agent accuracy
Atlassian has launched Code Context, a new feature within its Teamwork Graph that enables AI coding agents to access secure, multi-repository codebase understanding. Internal benchmarks show the feature improves agent accuracy by 44% and reduces token usage by 48% compared to agents operating without this context. The tool integrates code with organizational knowledge from Jira, Confluence, and other sources to help agents reason more effectively before generating or reviewing code. It is currently available in open beta for GitHub and Bitbucket users.

*this image is generated using AI for illustrative purposes only.
Atlassian Corporation (NASDAQ: TEAM) introduced Code Context, a new capability within the Atlassian Teamwork Graph, designed to provide developers and coding agents with secure access to source code across large-scale, multi-repository codebases. The update integrates organizational knowledge with codebase understanding to enhance the output quality of AI tools like Rovo and third-party coding agents.
The launch addresses a key limitation in current coding agent adoption: effectiveness depends not just on model intelligence but on the context available before action is taken. While agents can write and refactor code, they often lack access to architectural decisions, product strategy, and implementation discussions that explain why systems function as they do. Code Context creates a layer where this organizational knowledge connects with code, allowing agents to reason across both domains.
From code search to codebase understanding
Code Context establishes a secure, queryable representation of connected codebases, making this knowledge accessible across development workflows including IDEs, terminals, Jira, and Rovo Chat. Developers and agents can utilize exact search, natural-language queries, and semantic retrieval to locate relevant source code across multiple repositories.
Through the Teamwork Graph CLI, coding agents such as Cursor, Claude Code, and Codex can ground their work in relevant information before planning or generating code. Without this context, agents are typically constrained to code available on a developer’s local machine, risking changes that ignore broader system dependencies or usage patterns.
Performance benchmarks
Internal benchmarks indicate significant efficiency gains when agents are enriched by the Teamwork Graph. The data reveals two key performance improvements compared to agents operating without this context:
- Accuracy: Agents delivered 44% more accurate results.
- Token Efficiency: Agents used 48% fewer tokens.
These figures suggest that providing comprehensive context reduces the computational load required for verification while improving the reliability of the output. The system allows agents to combine code with signals from Jira work items, Confluence pages, Loom videos, and over 50 third-party connectors.
Security and governance
Code Context is designed with security controls to ensure agents only access information permitted to the specific user or authorized agent. Results are scoped based on existing permissions, ensuring that developers retain control over what ships. The goal is not unlimited access for agents, but rather right-sized access through a governed system.
Sanchan Saxena, senior vice president and head of product for Teamwork Collection at Atlassian, stated that organizations seeing success with AI are those providing it with the best context and right access. He noted that while frontier intelligence provides speed, context directs that speed toward impactful outcomes.
Availability
Code Context is currently rolling out via open beta to Atlassian customers, with support for GitHub and Bitbucket. Organization administrators can check availability under Rovo settings in Atlassian Administration.
How might the 48% reduction in token usage impact the overall cost structure of enterprise AI coding agents for large-scale organizations?
What competitive pressures could Atlassian face from other developer tool providers like GitHub or GitLab in integrating similar context-aware AI capabilities?
Could the reliance on the Teamwork Graph for cross-repository context create new vendor lock-in risks for enterprises currently using multi-cloud or hybrid development environments?
































