Thomson Reuters CTO Joel Hron explains how the company builds professional-grade AI for legal, tax, and compliance workflows using Claude. The company's approach, called Fiduciary-Grade AI, combines Anthropic's frontier models with Thomson Reuters' authoritative legal content, domain expertise from 2,700+ specialists, and rigorous evaluation infrastructure. CoCounsel Legal was rebuilt on the Claude Agent SDK to orchestrate hundreds of tools simultaneously, enabling tasks like legal research that once took dozens of hours to complete in minutes. Hron outlines four model requirements for high-stakes work: citation validation, reliable multi-step tool use, human-in-the-loop collaboration, and handling complex drafting tasks. He also shares a contrarian view on AI ROI, arguing teams should prioritize cultural mindset shifts before optimizing cost-per-task metrics.

7m read timeFrom claude.com
Post cover image
Table of contents
Evaluating models for legal workBuilding an agent-first productWhat knowledge work demands of a modelThe ROI of AIWhat's next
68 Impressions