Source time Edition date
ToolsLatent Space
Building AI for Lawyers - Jacob Lauritzen, CTO of Legora
为律师构建 AI——Legora 首席技术官 Jacob Lauritzen
Legora CTO Jacob Lauritzen's discussion of evals without traces and million-cell tabular reviews shows where legal AI still strains, and why the billable hour itself may be the thing that has to give.
Judgment
Law firm managing partners must decide whether to shift routine reviews to flat fees and absorb short-term revenue loss, because Legora's agent makes junior work faster while the billable hour rewards slowness.
Basis
Jacob Lauritzen said Legora cannot read customer traces because uploads are confidential and often client-owned; instead three engineers and three lawyers build RL environments, such as a 10,000-file data room for due diligence, and production classifies failures into 10 to 20 failure modes. He estimates models handle about 60% of legal work well, while complex analysis and editing across million-token contracts still need more intelligence. That shifts the buying question from model choice to workflow redesign: Legora runs on Azure because law firms had approved Microsoft via 365, and it post-trains open-weight models only for narrow tasks like legal research.
Unclear The source does not quantify how much flat-fee or free routine reviews cost firms, nor how Legora prices against saved billable hours.
Next Check Legora's next product release for a flat-fee review workflow or public eval heat-map disclosure.