The system functions by requiring the agent to review documents in a client matter to build a graph of entities and relationships, such as party names, dates and contract versions, before executing the task. Harvey plans to integrate this capability into its platform and share further experiments. Separately, the research organization DAIR.AI is applying similar structured data approaches to research-related tasks.
Tuesday, October 20, 2026
Story
Harvey Cuts Legal AI Agent Task Costs by 48% With Context Graphs
AI5 days ago
Legal AI agents saw their rubric pass rates climb by up to 30% in experiments conducted on the Legal Agent Bench (LAB), according to data from Harvey. The testing across 11 categories, including M&A diligence and contract negotiation, showed that agents using context graphs increased their success in matter counting from 53% to 83%, while firm knowledge and precedent list tasks rose to 80%. In one specific case, an agent's accuracy on a financing retrieval task jumped from 0% to 75% after adopting the graph system, which also reduced task costs by 48%.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.