
Fringe Legal Presents Bots @ Work How I built a law firm with AI agents with Helen Fan
10 snips
May 15, 2026 Helen Fan, AI and legal-tech practitioner who built an AI-native law firm, shares her OpenClaw experiment with two agents, Morgan and Cleo. She explains agent-to-agent argument reports, the technical and security pain of agent frameworks, and the Legal AI Value Stack. Hear why workflow orchestration and guardrails matter, and how AI-native and traditional firms are moving on different timelines.
AI Snips
Chapters
Transcript
Episode notes
Running A Two-Agent AI Law Practice
- Helen built OpenClaw Law LLP as an experiment with two AI agents, Morgan (senior associate) and Cleo (junior associate).
- Morgan handles strategy and client management while Cleo does legal research and drafts, and they can argue with each other producing argument reports.
Argument Reports Surface Hallucination Risks
- Agent-to-agent argument reports reveal points of disagreement and direct reviewers to where hallucinations or errors may occur.
- Helen uses these reports to focus her human review and believes they reduce hallucination risk by surfacing conflicting model views.
Expect High Maintenance With Open Source Agents
- Be prepared for instability and heavy troubleshooting when using open-source agent frameworks like OpenClaw.
- Helen warns agents break often, can loop endlessly, and require manual guardrail rules in agent.md and AgentMD to maintain safety.
