Context engineering: managing context quality for more efficient, effective AI
AI projects often fail due to poor context rather than model limitations, with 95% of enterprise pilots showing no business impact in 2025. Atlassian outlines how structured, connected, and relevant context can transform AI effectiveness.
In 2025, companies invested $581 billion in AI, yet 95% of enterprise pilots failed to improve business outcomes, according to an MIT study. The issue is not the AI models themselves but the context they rely on—every AI response depends on a packet of information assembled from an organization’s data. Unlike model quality, which is controlled by a few labs, context quality can be actively managed and improved by teams. Atlassian argues that refining context is a practical lever for making AI more efficient and effective in real workflows.
High-quality context enables AI to act like an experienced teammate rather than a new hire. For example, an on-call engineer at 2 a.m. might receive a generic runbook that lacks critical details, such as a recent service redeployment or a resolved incident’s workaround. Without this context, the engineer must search further, wasting time and tokens. Atlassian’s research identifies three drivers of context quality: completeness, connectedness, and relevance, which determine how well AI can reason over organizational knowledge.
Completeness ensures AI has access to all necessary information, including breadth across tools like Figma, Slack, and Google Docs, and depth in documents, comments, and edit histories. Connectedness links related pieces of context, such as a Figma link in a ticket or a PR reference in Slack, allowing AI to trace relationships like service ownership or deployments. Atlassian’s Teamwork Graph helps build this connectedness by mapping relationships organically as teams work, reducing token waste and improving response quality.
Relevance filters out outdated or sensitive information, ensuring AI prioritizes recent incidents over old ones and respects data classification. Atlassian’s tools, such as MCP server controls and Atlassian Guard, allow organizations to govern which Jira and Confluence content AI agents can access, enforcing permissions across projects or classification levels. However, the foundation of good context requires behavioral changes, such as labeling content, archiving stale material, and marking official pages, to ensure AI operates with accurate and secure data.