Context Graphs for Proactive Enterprise Agents
06:00 · July 11, 2026 · arXiv cs.AI RSS
Retrieval-Augmented Generation (RAG) and agentic frameworks have advanced enterprise AI considerably, yet agents remain fundamentally reactive: they wait for a human query before acting. This paper argues that genuine enterprise productivity gains require proactive agents: systems that surface relevant, actionable information to workers before they ask. We propose the Context Graph, a live relational data structure that models enterprise entities, their relationships, and state transitions over time. Built on this graph, we define a Delta Detection Engine that continuously monitors state changes, a Proactivity Scorer that ranks candidate insights by urgency, relevance, and persona-fit, and a Surfacing Layer powered by an LLM that delivers ranked notifications with grounded explanations. We formalize each component, derive a unified Proactivity Score function, and provide a complete end-to-end Python implementation using NetworkX and the Anthropic Claude API. Evaluation across three generic enterprise case studies (contract lifecycle management, engineering incident response, and sales pipeline hygiene) demonstrates that context-graph-driven proactivity achieves Precision@5 of 0.83, a false positive rate of 0.11, and reduces mean time to surface from 47 minutes (reactive baseline) to under 30 second.
Summary
This arXiv paper addresses the architectural limits of current enterprise AI agents, which remain reactive because they require an explicit user query before retrieving or acting on information. The authors argue that meaningful productivity gains depend on systems that can monitor enterprise state continuously and surface time-sensitive signals—such as an expiring contract or a stalled deal—without waiting for human initiation.
To enable this behavior, the work introduces the Context Graph, a directed, attributed, time-stamped multigraph that represents enterprise entities as nodes and their relationships as edges. Each node carries mandatory properties including a unique identifier, type, current state, timestamps, owner, and domain metadata, while edges are labeled with semantic relation types and weights. State changes on the graph are recorded as delta events that capture the entity, change type, old and new values, and logical clock, forming an immutable event log.
Three layered components operate on this structure. A Delta Detection Engine evaluates registered threshold rules against incoming deltas to identify candidate insights. A Proactivity Scorer then ranks those candidates according to a unified function that incorporates urgency, relevance, persona fit, and confidence. Finally, a Surfacing Layer uses an LLM to generate grounded natural-language notifications, applying deduplication and cooldown before delivery.
The paper supplies formal definitions for the graph schema and delta model, derives the Proactivity Score function, and includes a complete Python implementation built on NetworkX for graph operations and the Anthropic Claude API for notification generation. Evaluation on three representative scenarios—contract lifecycle management, engineering incident response, and sales pipeline hygiene—reports Precision@5 of 0.83, a false-positive rate of 0.11, and a reduction in mean time to surface from 47 minutes under a reactive baseline to under 30 seconds. The presentation targets researchers seeking reproducible technical detail rather than production deployment guidance.
Why it matters
High technical depth, novel proactive architecture, and complete reproducible implementation make it directly actionable for Dutch AI researchers and advanced enterprise practitioners developing agent systems.
