Uber
Uber's Legal Redlining Agent cuts average contract review time by over 20%
Uber Legal uses an AI add-in inside Microsoft Word that reviews counterparty edits to contracts and suggests responses for lawyers to accept, reject or change. Uber reports an over 20% reduction in average contract review time and 91% accuracy in AI-generated decisions.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
over 20%
Reduction in average contract review time since launch with the Legal team1
91%
Accuracy of AI-generated decisions1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Uber negotiates thousands of contracts a year. Much of the work is repeatable: similar clauses, the same edits from different clients, and standing legal positions. This work slows deals, onboarding and launches and takes lawyers' time away from complex judgment calls.
What they built
Uber built the Legal Redlining Agent, an add-in for Microsoft Word. A redline is an edit a counterparty makes to a draft contract. The agent finds those edits, works out what the other side intends, compares them with Uber's legal policies, and suggests whether to accept, reject or change each one. It also drafts comments back to the client and rates the risk of each change.
How they did it
Uber started with a simple search over legal playbooks, which gave unreliable results and the wrong tone. It then added a feedback loop that stores each lawyer decision and uses recent examples as guidance, a separate tone step written by the lawyers, drafting of counter-proposals, and a rules database for fixed positions. Lawyers review every suggestion. Uber piloted the agent with its Legal team in 2025.
Results
Since launch with the Legal team, Uber reports an over 20% reduction in average contract review time and 91% accuracy in AI-generated decisions. One lawyer is quoted as saying the agent saved real time, especially with comment suggestions.
What to copy
Pick a high-volume process with repeatable decisions. Meet users inside their current software. Put domain experts in charge of the wording, and collect their corrections from the start.
Lessons for your project
Build inside the tool people already use. Uber made the agent a Microsoft Word add-in because copying between tools would hurt adoption.
Let the experts own the wording. Uber's lawyers wrote the tone and style instructions and tuned them within days.
Capture every accept, reject or edit as feedback. Uber's agent improves from lawyers' decisions without retraining the model.
Keep lawyers as the decision-makers. Every suggestion is reviewed before it goes to the other side.
Caveats
- Both figures are Uber's own, from one engineering blog post co-written by its Legal team. No outside party confirms them.
- The post gives no baseline, sample size or period for the over 20% time reduction, and does not define how the 91% accuracy was measured.
- The post does not name the AI models used in this first-generation agent. It mentions exploring Claude Code and OpenCode for the future, which is not a statement of what runs today.
- The post gives no launch date. It says the agent was piloted with the Legal team in 2025 and built from early 2024, so the deployed date shows only the year 2025.
- Uber says the Legal team won an industry award on March 9, 2026 with the agent as part of its submission. This is not independent confirmation of the figures.
- No cost, team size or build cost is given.
Sources
- Scaling AI in Legal: Building Uber's Redlining Agent
Uber Blog · 2026-10-08 · Company blog · uber.com
Summarized in our own words from the sources above. Spotted an error? See corrections.