Arvind Jain emphasizes the significance of context graphs in the evolving landscape of enterprise automation. Context graphs are essential for understanding not only data but also the processes that drive a company’s operations. With nearly all enterprise activities digitized, context graphs illuminate how work gets done, enabling better automation through agents. These agents need a strong grasp of the enterprise context to function effectively. As AI takes on more complex tasks, understanding relationships within the organization becomes critical, especially since much of the work is done by individuals and small teams where processes often remain undocumented. Glean highlights that capturing the "how" of processes is more feasible than capturing the "why." The "how" refers to observable actions, such as document edits and Slack messages, which create a digital trail that can be analyzed. This approach allows organizations to infer underlying rationales based on patterns rather than attempting to document every human thought. Observability is key; it requires a comprehensive grasp of activities across various tools, ensuring that systems can collect and analyze data effectively. The challenge lies in separating meaningful signals from noise, especially since enterprise data sets are smaller and more private than typical internet-scale data. Building context graphs involves a technical investment and a combination of technologies, including connectors, indexes, and memory systems. These components work together to shift from raw data to actionable insights. As agents begin to operate within enterprises, their learning process becomes crucial. By capturing execution traces—how agents interact with tools and the outcomes of those interactions—organizations can refine their understanding of processes. The combination of structural knowledge and learned agent behavior creates a robust framework for automation, transforming how work is done in enterprises.