Building Legal Agents When You Can't See the Data — Jacob Lauritzen, CTO of Legora

Chapters

From the creator's chapter list. Click to jump.

1. 0:00Cold open
2. 0:40Meet Jacob Lauritzen, CTO of Legora
3. 1:05Starting out as underdogs in a Stockholm basement
4. 1:54How Legora grew from ChatGPT with Swedish law into a legal agent
5. 4:12The switch from assistant to agent
6. 5:23Which models Legora uses and why it added a model picker
7. 7:48Are today’s models good enough for lawyers?
8. 8:20What agents couldn’t do nine months ago
9. 9:18Testing the agent in simulated worlds and a 10,000-file data room
10. 10:07The evals team and Legora’s 100 legal engineers
11. 11:03When evals didn’t seem worth it and one release took citations from 70% to 99%
12. 12:18What legal practice could look like in five to ten years
13. 14:18Will the law firm partnership model survive?
14. 16:40What happens to paralegals and junior lawyers?
15. 18:00The billable hour and why earlier legal tech fell short
16. 19:39Where the money goes now
17. 22:44How Legora used its own product for acquisition diligence
18. 23:59How Swedish and American law firms differ
19. 26:12Product metrics when you can’t read the traces
20. 27:41Paraphrasing traces and narrowing failures down to 10 or 20 types
21. 28:42Why data privacy led Legora to run everything on Azure
22. 30:04Open-weight models and model arbitrage
23. 30:55Transcription models and why Legora doesn’t use ElevenLabs
24. 33:33The agent, Tabular Review, projects, and the Word add-in
25. 34:24How a lawyer uses Tabular Review
26. 35:51Why a spreadsheet works and what you learn from the first 10 rows
27. 37:53How the legal intake bot works in Slack
28. 39:11How agents create lists and lists start agents
29. 40:15Building products on those same pieces and acquiring Wexler
30. 42:37What forward-deployed engineers do at Legora
31. 43:52How legal engineers and forward-deployed engineers onboard a big client
32. 46:43What’s next for the forward-deployed team over the coming year
33. 47:24Why Legora resisted building APIs
34. 49:26Can the agent be the API?
35. 50:12The 300 Spartans slide and the plan to stay at 20 engineers
36. 51:32The bets nobody asked Legora to make
37. 53:05Post-training open-weight models for legal research
38. 54:05Search, retrieval, and late interaction
39. 54:51What makes a good legal agent harness and why drafting depends on search
40. 56:25Which model writes the best draft?
41. 57:03Hallucinations and Legora’s citation system
42. 57:41Do lawyers check the citations?
43. 58:19MapReduce or one big context window for a million-token contract?
44. 1:00:16Getting anonymized examples and the idea of buying a law firm for its data
45. 1:01:53Why Jacob advises against default embeddings
46. 1:02:44Will lawyers work in Legora or through Claude?
47. 1:03:26Will the model labs cut Legora off?
48. 1:04:24Law schools and scholarships for computer science majors
49. 1:05:28Hiring in New York, Stockholm, and London
50. 1:06:09Why most work eventually becomes legal work