{"slug": "building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora", "title": "Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora", "summary": "Legora CTO Jacob Lauritzen said the legal-tech company builds and tests its legal agent in simulated environments modeled on customers' work because it cannot read any trace of what the agent does for a lawyer, since the data belongs to law firms' clients. Legora's evals team, working with three engineers and three lawyers, narrows failures down to 10 or 20 types, and Lauritzen cited a citation problem where the model was correct 70% of the time and reached 99% after one release without Legora changing anything. Lauritzen also estimated that today's models already do about 60% of legal work well, and Legora now runs everything on Azure for data privacy and employs 100 legal engineers after once arguing 20 would suffice.", "body_md": "Legora can’t read a single trace of what its agent does for a lawyer. The data belongs to the law firm’s clients. How do you build and test an agent when you can’t see its work?\n\nJacob Lauritzen is the CTO of Legora, one of the fastest-growing companies in legal tech history. The team started in a Stockholm basement with ChatGPT, Swedish law, and a search tool. Today, a lawyer can use Legora to draft and edit Word documents, review 10,000 contracts in a single table, or run due diligence. The agent also coordinates work between lawyers, clients, and other agents. When Legora made its own acquisitions, the team used it for diligence.\n\nWhat we talked about\n\n• Why Legora resisted adding a model picker for years, and why it finally added one\n\n• Jacob’s estimate that models already do about 60% of legal work well, and why the rest takes a smarter model\n\n• How three engineers and three lawyers build test environments modeled on customers’ work, much like the “worlds” used in reinforcement learning\n\n• A citation problem where the model got it right 70% of the time, then 99% after one release. Legora didn’t have to change anything.\n\n• Why Legora runs everything on Azure and why it doesn’t use ElevenLabs\n\n• Tabular Review, the million-cell spreadsheet the industry copied down to the name\n\n• How intake works, how agents create lists, and how a list can start another agent\n\n• What legal engineers and forward-deployed engineers each do, and who gets to write code\n\n• The 300 Spartans slide Jacob used to argue that 20 engineers would be enough forever. Now there are 100, and he still feels short-handed.\n\n• Why Jacob finds Fable and Opus too chatty for drafting and likes Astra’s concise writing\n\n• Why drafting a long contract is a search problem\n\n• Whether model labs will cut off legal tools the way they cut off coding tools\n\n• What happens to law firm partnerships, the billable hour, and jobs for junior lawyers\n\nLegora is hiring engineers in New York, Stockholm, and London.\n\nChapters\n\n0:00 Cold open\n\n0:40 Meet Jacob Lauritzen, CTO of Legora\n\n1:05 Starting out as underdogs in a Stockholm basement\n\n1:54 How Legora grew from ChatGPT with Swedish law into a legal agent\n\n4:12 The switch from assistant to agent\n\n5:23 Which models Legora uses and why it added a model picker\n\n7:48 Are today’s models good enough for lawyers?\n\n8:20 What agents couldn’t do nine months ago\n\n9:18 Testing the agent in simulated worlds and a 10,000-file data room\n\n10:07 The evals team and Legora’s 100 legal engineers\n\n11:03 When evals didn’t seem worth it and one release took citations from 70% to 99%\n\n12:18 What legal practice could look like in five to ten years\n\n14:18 Will the law firm partnership model survive?\n\n16:40 What happens to paralegals and junior lawyers?\n\n18:00 The billable hour and why earlier legal tech fell short\n\n19:39 Where the money goes now\n\n22:44 How Legora used its own product for acquisition diligence\n\n23:59 How Swedish and American law firms differ\n\n26:12 Product metrics when you can’t read the traces\n\n27:41 Paraphrasing traces and narrowing failures down to 10 or 20 types\n\n28:42 Why data privacy led Legora to run everything on Azure\n\n30:04 Open-weight models and model arbitrage\n\n30:55 Transcription models and why Legora doesn’t use ElevenLabs\n\n33:33 The agent, Tabular Review, projects, and the Word add-in\n\n34:24 How a lawyer uses Tabular Review\n\n35:51 Why a spreadsheet works and what you learn from the first 10 rows\n\n37:53 How the legal intake bot works in Slack\n\n39:11 How agents create lists and lists start agents\n\n40:15 Building products on those same pieces and acquiring Wexler\n\n42:37 What forward-deployed engineers do at Legora\n\n43:52 How legal engineers and forward-deployed engineers onboard a big client\n\n46:43 What’s next for the forward-deployed team over the coming year\n\n47:24 Why Legora resisted building APIs\n\n49:26 Can the agent be the API?\n\n50:12 The 300 Spartans slide and the plan to stay at 20 engineers\n\n51:32 The bets nobody asked Legora to make\n\n53:05 Post-training open-weight models for legal research\n\n54:05 Search, retrieval, and late interaction\n\n54:51 What makes a good legal agent harness and why drafting depends on search\n\n56:25 Which model writes the best draft?\n\n57:03 Hallucinations and Legora’s citation system\n\n57:41 Do lawyers check the citations?\n\n58:19 MapReduce or one big context window for a million-token contract?\n\n1:00:16 Getting anonymized examples and the idea of buying a law firm for its data\n\n1:01:53 Why Jacob advises against default embeddings\n\n1:02:44 Will lawyers work in Legora or through Claude?\n\n1:03:26 Will the model labs cut Legora off?\n\n1:04:24 Law schools and scholarships for computer science majors\n\n1:05:28 Hiring in New York, Stockholm, and London\n\n1:06:09 Why most work eventually becomes legal work\n\nGuest — Jacob Lauritzen, CTO of Legora\n\nCo-host — swyx, Latent Space\n\nHost — Basil Chatha, Forward Deployed", "url": "https://wpnews.pro/news/building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora", "canonical_source": "https://latent.space/p/building_legal_agents_when_you_can_t_see_the_data_-_jacob_lauritzen_cto_of_legora/", "published_at": "2026-11-02 22:54:03+00:00", "updated_at": "2026-11-02 23:19:39.575488+00:00", "lang": "en", "topics": ["ai-agents", "ai-products", "large-language-models", "ai-safety"], "entities": ["Legora", "Jacob Lauritzen", "Azure", "ElevenLabs", "Wexler", "ChatGPT", "Stockholm", "New York"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora", "markdown": "https://wpnews.pro/news/building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora.md", "text": "https://wpnews.pro/news/building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora.txt", "jsonld": "https://wpnews.pro/news/building-legal-agents-when-you-can-t-see-the-data-jacob-lauritzen-cto-of-legora.jsonld"}}