Legora, one of the fastest-growing legal tech startups, builds that draft documents, review contracts, and run due diligence for law firms—but cannot read a trace of the work because client data belongs to lawyers and their clients. In a technical interview, CTO Jacob Lauritzen explains how the team tests agents by building synthetic environments modeled on real customer workflows, similar to reinforcement learning "worlds." He describes why Legora resisted multi-model support for years before adding it, estimates that AI already handles roughly 60% of legal work reliably, and discusses challenges like citation accuracy that sometimes improve unexpectedly after model updates.
This is an AI-written summary. The original reporting belongs to @LatentSpacePod. How we write these · Spot a mistake?