Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Uber's legal AI agent cuts contract review time by 20%

Average contract review time fell more than 20%.

Uber's Legal Redlining Agent has reduced average contract review time by more than 20%, achieving 91% accuracy in AI-generated decisions, according to the company1,2.

The agent integrates directly into Microsoft Word, where it analyzes proposed contract changes, assesses legal risks, and recommends policy-aligned responses. It identifies external-party edits, interprets their intent, and compares changes against legal policies, suggesting whether to accept, reject, or modify clauses and generating comments that explain its reasoning. Lawyers remain responsible for reviewing and approving all recommendations.

The system is built on LangGraph, OpenSearch, and Redis. It combines retrieval-augmented generation, historical lawyer feedback, and deterministic policy rules. OpenSearch stores vectorized legal rules and historical feedback, while Redis maintains session state for follow-up questions. A deterministic rules engine can override model recommendations. The front end is a React and TypeScript Word add-in backed by a Python service orchestrated through LangGraph.

Uber is exploring agentic harnesses, skills, legal ontologies, and knowledge graphs as extensions to the system.

ANALYSIS The architecture is notable for layering deterministic rule overrides on top of RAG-based generation, a pattern that lets legal teams enforce hard policy constraints while still benefiting from language-model flexibility. The human-in-the-loop design, where lawyers approve every recommendation, keeps the agent in an advisory role rather than an autonomous one.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.