Uber Legal AI Agent

Uber's AI Agent Saves Lawyers 20% Time on Contract Reviews

Uber's legal team has built an AI agent that cuts contract review time by over 20%, with 91% accuracy on AI-generated decisions.

3 min readInfoQ
Uber's AI Agent Saves Lawyers 20% Time on Contract Reviews

Uber's Legal Redlining Agent is a concrete example of what happens when you stop treating AI as a novelty and start treating it as a practical tool for knowledge work. The system, which reviews contract changes in Microsoft Word using retrieval-augmented generation, historical lawyer feedback, and deterministic policy rules, has already delivered more than 20% lower average contract review time with 91% accuracy on AI-generated decisions. That is not a theoretical promise, it is a measurable outcome from a team that built on LangGraph, OpenSearch, and Redis. For anyone who has watched legal teams drown in redlines, this is the kind of progress worth paying attention to.

What makes this development stand out is the deliberate architecture underneath it. Uber did not throw a large language model at the problem and hope for the best. They combined RAG with historical feedback loops and deterministic rules, creating a system that learns from actual lawyer decisions rather than generic training data. This mirrors a broader shift we are seeing across enterprise tooling: the move from black-box AI to governed, auditable systems. It is the same principle behind Governed AI brings trusted autonomy to mission-critical data work, where the focus is on combining human oversight with machine speed rather than replacing judgment outright. Both approaches acknowledge that accuracy matters more than speed when the stakes involve legal obligations.

The 91% accuracy figure is impressive, but the open question is what happens to the remaining 9%. Uber reports that the agent is already being extended toward agentic harnesses, skills, legal ontologies, and knowledge graphs. That suggests they are treating this as a foundation, not a finished product. The trajectory matters because contract review is just one domain where structured knowledge meets repetitive human work. We are seeing similar patterns in database management, where DocumentDB 1.0 Brings Open Source MongoDB Compatibility to PostgreSQL demonstrates how legacy workflows can be modernized without forcing teams to abandon their existing tools. The common thread is pragmatism: meet users where they are, then give them a clear path forward.

The specific takeaway here is that AI agents in enterprise settings succeed when they respect existing workflows rather than trying to replace them. Uber's system lives inside Microsoft Word, learns from the lawyers who use it, and reports its confidence levels so humans can override when necessary. That is not a flashy vision, it is a practical one. The detail to watch is how Uber handles the knowledge graph layer. If they can encode legal ontology in a reusable way, the same architecture could scale to compliance review, regulatory filings, or procurement. That is where the real transformation begins.

From InfoQ

Uber’s Legal Redlining Agent reviews contract changes in Microsoft Word using RAG, historical lawyer feedback, and deterministic policy rules. Built with LangGraph, OpenSearch, and Redis, the system reports more than 20% lower average contract review time and 91% accuracy in AI-generated decisions. Uber is exploring agentic harnesses, skills, legal ontologies, and knowledge graphs.

Read the original at InfoQ