Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
November 2, 2026
AI Summary
5 min readThe Agent Lab for Law: How Legora Builds for a Profession That Can't Share Its Data
When Jacob Lauritzen's team at Legora first built an AI assistant for lawyers, the simplest version was "ChatGPT with a Swedish law search tool in it." That was 2023. Today, Legora's agent can fan out across a million documents, extract structured data into spreadsheet-like tables, draft complex contracts from precedent, and orchestrate workflows through a custom task system — all while operating under constraints that would make most AI engineers wince: the company can't see what its users are doing with the product.
"Here's an agent and it can do so many amazing things and I have no idea what people are actually doing," Lauritzen says. All client data is confidential, often belonging to the law firm's clients, not the firm itself. Legora can't inspect traces, can't read prompts, can't look at outputs. This forces the company to build its entire product intelligence system differently than a coding tool or a consumer app would.
The Evals Problem When You Can't See the Data
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:00) **Opening & The Underdog Origin** - Jacob Lauritzen introduces himself and contrasts Legora's humble, European underdog origin with the boldness needed to compete in the US market.
- 2 (02:13) **Product Evolution: From Assistant to Agent** - A walkthrough of Legora's product history, starting with legal research and expanding into drafting, analysis, and workflow orchestration.
- 3 (05:24) **Model Selection & The "Model Picker"** - The team uses multiple models depending on the task and has reluctantly introduced a model picker for users despite preferring to abstract model choice away.
- 4 (08:44) **Sub-Agent Delegation & The Evals Team** - How Legora tracks model performance across legal sub-domains using a dedicated evals team composed of both engineers and lawyers.
- 5 (11:54) **The Real Work: Product, Privacy, and the Future of Law** - The core job is not just exposing model capability, but building a complete product with audit logs, data retention, and a vision for the future of legal work.
- 6 (16:49) **The Impact on Junior Lawyers & Paralegals** - A debate on whether junior roles are being eliminated or transformed, drawing parallels to the software engineering industry.
- 7 (22:44) **Dogfooding M&A with Legora** - Jacob describes how Legora used its own product to perform due diligence for its acquisition of Cura.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
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? Jacob Lauritzen is the CTO of Legora, one of the fastest-growing companies...