Paper proposes Context Graphs to make enterprise agents proactive instead of reactive

Current enterprise AI agents and retrieval-augmented generation (RAG) systems are fundamentally reactive: they wait for explicit human input before initiating action. This limits their utility for time-sensitive workflows where delays between event occurrence and human awareness can be costly.

The paper introduces the Context Graph, a live relational data structure that models enterprise entities, their relationships, and state transitions over time. This structure enables continuous monitoring of changes in enterprise state rather than waiting for user queries.

To operationalize the Context Graph, the authors define three components: a Delta Detection Engine that continuously monitors for state changes, a Proactivity Scorer that ranks candidate insights by urgency, relevance, and persona-fit, and a Surfacing Layer powered by a large language model that delivers ranked notifications with grounded explanations.

The authors formalize each component and derive a unified Proactivity Score function to prioritize which insights to surface to which users. They also provide a complete end-to-end Python implementation using NetworkX for graph operations and the Anthropic Claude API for the Surfacing Layer.

Evaluation across three generic enterprise case studies—contract lifecycle management, engineering incident response, and sales pipeline hygiene—reports Precision@5 of 0.83 and a false positive rate of 0.11. The system reduced mean time to surface actionable information from 47 minutes in a reactive baseline to under 30 seconds in the proactive setting, representing a greater than 90% reduction in latency.