Context graphs are becoming essential for enterprise AI, helping systems grasp the actual processes and relationships that define how work gets done. By focusing on the "how" of work—tracking actions and collaborations—they enable organizations to automate complex tasks that typically rely on undocumented knowledge or "tribal knowledge." This shift allows AI agents to learn from existing workflows, which is crucial as more enterprise operations are digitized across various tools like communication platforms and project management systems. Building effective context graphs is technically challenging. It requires a robust infrastructure that includes connectors for real-time observability, capturing activity data, and understanding the relationships within the organization. For instance, a context graph can track interactions across multiple applications, like Salesforce, Google Docs, and Slack, to create a comprehensive view of workflows. The goal is to analyze these interactions to infer higher-level tasks and projects, even if they aren't explicitly labeled. Currently, Glean's task understanding accuracy is around 80%, a testament to the complexity involved in distinguishing meaningful signals from noise in an enterprise environment. As AI takes on more complex tasks, understanding context becomes more critical. Glean emphasizes that simply indexing unstructured data isn't enough; systems must also capture the nuances of relationships within the organization. The article argues that while context graphs are gaining attention, they function as part of a broader technological stack that includes connectors, indexes, and memory systems. This combination enables a deeper understanding of processes, allowing agents to learn and adapt over time based on execution traces.