4 tips for building better AI agents that your business can trust
Joel Hron's framework includes human-in-the-loop validation and expert collaboration for reliable agent deployment.
Joel Hron, CTO at Thomson Reuters Labs, has distilled four critical lessons from the company's extensive work with AI agents, including the legal research tool Westlaw Advantage and the Deep Research agent. The first pillar is rigorous measurement: Hron stresses the need to define "what good looks like" through a combination of public benchmarks, proprietary internal evaluations, and, crucially, human expert validation before any product ships. This human-in-the-loop approach provides the confidence needed to deploy reliable systems.
Second, Hron advises tightly coupling agent awareness with user experience, creating a common language and interface for human-AI collaboration. The third tip is to foster close collaboration between AI developers and domain experts, ensuring the technology is built around real user needs. Finally, he champions a culture of continuous experimentation, allowing teams to test ideas rapidly using automated evaluations while maintaining a high bar for final quality through expert oversight.
- Define success with a mix of public benchmarks, internal automated evaluations, and mandatory human expert validation before shipping.
- Create a common language and interface for seamless human-AI collaboration, treating agents as team members.
- Foster tight collaboration between AI developers and domain experts to ground technology in real user workflows.
Why It Matters
Provides a proven, enterprise-grade framework for deploying reliable AI agents that augment, rather than replace, professional expertise.