Imagine hiring a developer, showing them one folder and asking them to change a system spread across several teams. Even an excellent engineer would start by asking questions. Atlassian’s August introduction of Code Context makes a related argument about AI coding agents: capability alone does not provide the missing organisational brief.
Announced on August 12, Code Context connects codebase understanding to Atlassian’s Teamwork Graph. The company describes permission-aware access to source code and related organisational material, with retrieval across repositories. Administrators must enable code indexing before teams use the capability.
Atlassian said its internal benchmarks produced more accurate results with fewer tokens when agents had the additional context. These are vendor-run benchmarks, not a guarantee for an individual development team. The announcement also explicitly keeps responsibility for reviewing and shipping a change with developers.
GFT analysis: the interesting business problem is the cost of explaining a company to its software. A repository records implementation, while a decision document may explain why an apparently simpler alternative was rejected. Connecting those two kinds of knowledge could reduce repeated investigation. It could also make old or contradictory documentation more consequential.
That creates a useful test for teams considering contextual AI tools. Choose a real task that crosses a system boundary, agree on the correct answer in advance and compare the amount of correction required. A polished explanation is less meaningful than whether the proposed change respects the actual architecture and avoids new work elsewhere.
Permission boundaries deserve the same attention as accuracy. Information available to an organisation is not necessarily appropriate for every agent or employee. A good evaluation therefore asks both whether the tool finds the missing context and whether it refuses material outside the user’s access. The strategic opportunity is not simply to give AI more information. It is to give it the right information with a traceable reason to trust the result.