AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Building AI for Lawyers - Jacob Lauritzen, CTO of Legora

Collected Nov 3, 2026

Legora CTO Jacob Lauritzen sat down with Latent Space shortly after the legal AI company announced a $550 million Series E at an $8.8 billion valuation. Legora, which describes itself as a collaborative AI platform for lawyers, grew out of a Stockholm basement product that amounted to ChatGPT plus Swedish law plus a search tool. Lauritzen had already been through Y Combinator with his own startup when he helped Max Junestrand with the application for Leya, now Legora; he later joined as head of engineering and is now CTO.

The numbers are striking for a company less than two years past launch. Legora passed $100 million in ARR in April, less than 18 months after launch, and now counts more than 1,800 customers. It raised $550 million in March at a $5.55 billion valuation, then another $550 million in October at $8.8 billion. Its latest acquisition, Wexler — a London startup that extracts and verifies facts across large litigation document sets — was its fifth purchase this year. The show notes put ARR above $200 million at the time of the Series E announcement.

The product today drafts and edits Word documents, reviews thousands of contracts at once, and coordinates work among lawyers, their clients and other agents. It is built on a small set of primitives: projects that hold files, the agent itself, Tabular Review and lists. Tabular Review treats each row as a document and each column as a question — drop in 10,000 employment contracts and ask for termination periods, IP assignment clauses and severance terms. It supports up to a million cells but starts with just 10 rows so users can fix their prompts before running the full set. Lists are a Notion-style database with triggers and automations, so an agent can create and read a list and a change to a list can start an agent. Intake is a Slack bot that turns a request into a ticket on a list. Legal engineers configure new products on these primitives without writing code, and Wexler's fact intelligence product runs on the same building blocks.

The most interesting engineering problem Lauritzen describes is evaluation. Most agent teams improve by reading traces of real usage, but Legora cannot read any: nearly everything users upload is confidential, and much of it belongs to the firm's clients rather than the firm. So a team of three engineers and three lawyers builds reinforcement-learning environments that simulate the work customers actually do. One example is a 10,000-file data room where the agent runs due diligence and must produce a report that catches the right red flags. Results are plotted on a heat map by practice area and task, because a new model often gets better at some kinds of legal work and worse at others. In production, Legora classifies failures into 10 to 20 known failure modes and runs cheap LLM-based checks behind feature flags. This is a plausible model for any regulated, data-sensitive vertical: since you cannot observe real traffic, you synthesize it and measure per-task rather than in aggregate.

The same privacy constraints shaped Legora's infrastructure. Everything runs on Azure because, when the company started, law firms would not accept AWS or GCP, and Microsoft was already an approved vendor through Office 365. That is a useful reminder that in enterprise sales, procurement precedent often decides the cloud before any technical argument does.

On models, Legora resisted a model picker for a long time before adding one. Lauritzen's position is that a lawyer should not have to decide whether a contract is a high-effort or mid-effort task; routing is Legora's job, balancing quality, cost, latency, governance and regional availability. He estimates models already handle roughly 60% of legal work well, which makes those tasks a cost and latency problem rather than an intelligence problem, while complex analysis and editing across million-token contracts still need more capable models. Legora has begun post-training open-weight models for narrow tasks such as legal research, with no plans to train its own frontier model. For drafting, he finds Fable and Opus too verbose and Astra more concise. To combat hallucination, everything the agent writes carries a citation that highlights a bounding box in the source PDF, layers down. One anecdote stands out: a citation problem that sat at roughly 70% was fixed to about 99% by a model release, without the team doing anything.

The go-to-market model is unusual. Lauritzen once showed the team a slide of the 300 Spartans and declared Legora would stay at 20 engineers forever; it now has about 100 and he says that still feels too small. The company hires legal engineers — lawyers who spent three or four years in big law, often with a CS degree, roughly 100 of them — who configure the product the way a power user would and write no code. Forward deployed engineers came later, starting about nine months ago in New York with Di Qi, a two-time YC founder who worked on search at Meta. FDEs write code into production, build integrations into a firm's CRM or document management system, and feed what they learn back into the core product; they still do not touch client data. When a large engaged client signs, legal engineers are assigned and an FDE is added if the project warrants it. Within a year, Lauritzen expects a rotating group of FDEs who sit with a client for a couple of months, build what is needed and move on.

On the business model, he notes that legal tech failed for years because hourly billing meant faster work meant less revenue. What changed is pressure on firms: in the US, routine work such as small reviews is shifting to flat project fees, and some firms do it free to retain the client, while revenue moves toward M&A, litigation and client-facing time. He disputes the idea that paralegals and junior lawyers disappear, saying a good junior is worth more today than two years ago because they accomplish so much more.

Looking ahead, he expects large enterprises to reach Legora through tools like Claude Cowork in about two years while law firms work inside Legora directly. He does not expect model labs to cut Legora off the way they have coding companies, and argues that for a $400 million M&A transaction, buyers will still want Legora.

Why it matters: teams building agents in regulated, confidentiality-bound domains face the same bind Legora does — no trace access — and its simulated-environment plus failure-taxonomy approach is a concrete pattern to borrow. Inference: the legal-engineer and FDE split also suggests vertical AI companies will compete on service delivery and workflow orchestration as much as model quality, which raises the cost of every enterprise deal and rewards firms that can productize what field teams learn.

Read at Latent Space

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Evals without traces, million-cell tabular reviews, and cracks in the billable hour as a business model