← back
Agents need more than a chat - Jacob Lauritzen, CTO Legora
Takeaway
Long-horizon vertical agents need verifiability proxies, task decomposition, guardrails, and continuous human steering — chat is the wrong interaction model.
Summary
- Legora (legal vertical AI, 1000+ customers, 50+ markets) argues complex agentic work needs more than chat — planning and reviewing are now the bottleneck, not doing.
- Applies the 'verifier's rule' (Jason Wei): verifiable tasks get solved by AI; legal tasks span the spectrum (definition checks easy, contract drafting and litigation strategy hard to verify).
- Trust levers: pull tasks toward verifiability (golden-contract similarity proxies), decompose into sub-tasks (formatting, definition linting), and apply guardrails like restricted file/web access.
- Control levers: chat alone gives only root-level steering; planning steers up front but blind to discoveries mid-task; skills, hooks, and live mid-execution intervention give finer control.
- Future of agent UX is multiplayer document/canvas surfaces where humans steer continuously, not turn-based chat.
agentslegalverifiability
Original description
Jacob Lauritzen is CTO at Legora, the fastest growing legal tech startup in history. Speaker info: - https://www.linkedin.com/in/jacob-lauritzen/ - https://github.com/Jacse