Legora's CEO on Building Legal AI That Survives the Next Model Release

Jacob Effron introduces Max as the CEO and co-founder of Legora, a company applying AI to the legal industry and working with many top law firms around the world. He notes that Legora has raised over $100 million and is one of the fastest-growing AI applications. Max then recalls telling the board at the first meeting after a $35 million raise that the company would not sell for the next four or five months. The hosts frame this as an unusual but mature decision for a first board meeting.

The hosts frame the episode around Max's view of the legal space, product prioritization as models improve, what to build versus leave to model providers, and expansion across global markets. Max joins from Stockholm, and the conversation starts with where AI meets law today. Legora's work with top law firms is the backdrop for those questions.

Max says that when Legora started in 2020, nothing in legal AI was working. The team used early BERT models from Google, which were decent in English but "horrendously bad" in Swedish. GPT-3.5 became the paradigm starter, and the field moved from full-on experimentation into implementing work that actually holds up. By today, he says, legal AI is no longer just empty queries against a dataset.

For due diligence, lawyers no longer need to go physically into a data room or use FTP; they can put documents into Legora, state what they want to find, and the system finds it. Legora then generates a report from those findings. Max describes the shift as giving an agent access to tools so it can plan, execute the plan, and return an end-to-end work deliverable that is actually usable.

When asked where the field goes in five to ten years, Max says better models are only one piece. Legora gets the most leverage from surrounding frameworks such as function calling, tool calling, and MCP. Legal software had been fragmented across translation, document comparison, search, and review tools, and those capabilities are now being baked together.

Max describes legal work as a complexity ladder: at the bottom are simple tasks like finding data and extraction, and at the top are complex tasks like drafting a share purchase agreement. Legora has already started to fully automate much of the bottom quartile of that graph and is moving upward. The key question for law firms is where their specific expertise and context add the most leverage, versus where off-the-shelf LLMs are plainly good enough.

Max did not come from law; he was an engineer first. When he arrived, he noticed that very little software had been developed for the space and that industry incentives did not always reward efficiency. In-house counsel tend to handle repeated work such as NDA reviews and MSAs to control business risk, while law firms handle more one-off or complex projects with less precedent. The work can be broadly stitched into reviewing, reading, drafting, writing, and researching.

Legora emerged as a platform serving the whole wall-to-wall set of legal needs rather than a point solution. Max says law firms want to lead in the new paradigm, and firms that do not lean in risk failing to upskill their teams and facing client pressure. Clients are starting to use these tools internally, and executives publicly say they will go AI-first and demand proof of efficiency before granting new headcount. Max cites the Fiverr CEO mentioning Legora for its legal stack, and he notes that legal software is about a $20 billion market while legal services is about a $1 trillion market.

Jacob raises the hourly-billing problem: if software makes a lawyer 50% more efficient, the lawyer bills 50% fewer hours. Max agrees but points to writeoffs and price pressure that are not always visible from outside the industry. Due diligence used to be expensive and is approaching the point where clients are unwilling to pay for it; large clients may pay for board advisory but not contract review. He notes that large American firms are already outsourcing this work because it is not profitable to staff an associate billing $800 an hour. He calls it a prisoner's dilemma, because if one competitor moves, low differentiation on particular tasks forces everyone to follow.

Max describes Legora's start in the Nordics as a small fish eating its way into becoming a crocodile in a smaller pond before jumping across the Atlantic. The company recently opened in New York, and Europe's fragmentation meant its initial market was small. During Y Combinator, the team was pushed on how it would move beyond Sweden; Max said they would win Sweden first and then expand. As a fast second mover, Legora could observe what was working and what was not.

At the time, many competitors focused on training their own LLMs and presenting themselves as AI research companies. Legora had a $50,000 angel round and decided it did not have the money, time, or energy for that path. Instead, it focused on the application layer and building something people were excited to use. Coming from a non-legal background also forced the team to be humble and attentive to what clients were saying about law firm-client relationships.

Jacob relays another founder's observation that starting in the Nordics allowed a broader product because the US market would have forced a narrower focus. Max says this explains what Legora did: many legal tech companies solve a narrow problem like drafting or contract review, but Legora wanted every serious lawyer to use it the way great designers use Figma. Competitive pressure in the US might have made that ambition harder to sustain.

Serving enterprise clients in the local market made Legora enterprise-ready before entering new markets. Law firms are well connected, and a large Swedish firm often has partner or friendly firms in France, Germany, Spain, and the US, so working with a few firms produced referrals. Max adds that teaching lawyers to use the tools took far more work than expected. Jam, who came from Baker McKenzie and had run a GenAI rollout, said traditional software rollouts were thrilled with 5-10% adoption, but Legora deployments increasingly hit 70-80% adoption as lawyers actively ask the innovation team for access.

Max argues that model providers should also be viewed as platforms and software companies. OpenAI, Anthropic, and Google are not only pushing model capability but also building products that connect with other systems: Claude connects to other systems, Gemini sits on top of Google Workspace, and the labs shape how software interacts. His rule is that if the AI labs are likely to build something and make it available to builders, Legora should not build it. The goal is to build boats so that when the model tide rises, all the functionality gets better.

Legora's playbooks feature is an example of value that may be temporary. Users give Legora a set of negotiation rules and fallbacks, such as 20 rules for negotiating an NDA or MSA, and Legora goes through them one by one to make high-quality edits. Max says that if models become good enough to take a Word or Excel playbook, cross-reference another NDA, and return all the red lines, the feature may not need to be built that way. It adds value today but may be unnecessary in five years.

The same tension applies to multi-step workflows. Traditional node-based builders connect the output of one block to the input of the next and require technical builders to set up workflows. Max says LLMs with proper planning and tool use can create their own plan on the fly, so the user only needs to provide an instruction and the tools at its disposal. He acknowledges there is no perfect way to predict model capabilities even tomorrow, let alone six or twelve months out.

Legora built citations early because lawyers must see the source material and specific text chunks behind an LLM response. When Anthropic released citations in its API, Max's view was that if the LLM providers give that for free, Legora will deprecate its own code. His general framework is: if somebody else is doing it, use that. The hardest part of building is prioritization, because there are a hundred high-value things to do, and choosing five that fit a cohesive platform is what prevents a Frankenstein product.

Legora runs a seat-based model because it is easy to buy and predict, but Max says one user recently racked up $10,000 in LLM costs. He can imagine a platform fee plus a usage element over time, though that depends on how LLM providers develop their pricing. The early hope was that LLM prices would keep falling, allowing Legora to take lower margins early and improve them later. That has not played out cleanly because models are getting better and more expensive at the same time.

Max says o3 is incredibly good at legal work but also very expensive. The practical question becomes whether a task needs "the bazooka or a handgun." Legora increasingly runs classification algorithms and model pickers that choose the best model for the task, rather than assuming one model fits everything. He notes that a model that is expensive today may be cheap next year and also last year's capability.

MCP is the most interesting infrastructure development for Legora because it lets the product call outside tools. Legora can give its agent tools such as redlining a document, but the universe expands when clients can provide their own tools: a client-specific CRM, a knowledge database, templates, or the ability to push notifications to client email. On moats, Max points to Figma as a system of record where designers, PMs, marketers, and clients collaborate; Legora already integrates with customer data and external databases for case law, legislation, and regulation. He notes that the average big-law lawyer spends about 80% of their time in Word, Outlook, and document management systems, and he expects that to change.

Max says Legora has set up thousands of evals on Braintrust, which makes it easy to test a new model as soon as it is available. Every model Legora uses also goes through security, privacy, and legal review so it conforms with data processing agreements, which matters when working with law firms. For multi-step workflows, Legora has evals for each step and evals for the complete end-to-end product, and it tests permutations of different models in different places. High-quality reasoning models are very good, Max says, but they are hard to price.

Max recalls being accepted early into Y Combinator's winter batch in August. In an unconventional move he does not recommend, he took out a loan against the YC investment because Legora had a Swedish corporation that needed to flip to Delaware, a process that would take about three months. The team needed money immediately to hire four more engineers. At YC, everyone seemed to be working with AI, either on vector databases and Postgres layers or on application-layer products.

Max says he did not participate much in YC because he was handling commercial work. He bought a ring light for his laptop and sold to Europe between 1 a.m. and 10 a.m. for five or six days straight, then crashed and returned to Stockholm. Legora was commercially focused from the beginning and took to heart the advice to launch as soon as it had something. Its first product was essentially a private, compliant ChatGPT with better RAG on customers' documents and Swedish legislation, and the bar rose every week.

Max says lawyers get one chance: if an attorney runs a query and it does not work, they fall off and reactivation is hard. In the beginning, Legora had to solve reliability, chunking, RAG quality, and the ability to serve thousands of users a day before onboarding large firms. As Legora moved from large European firms to some of the largest firms in the world, expectations rose and the product became a system of record relied on for end-to-end deliverables, with immediate phone calls and Slack messages when something breaks. The company moved from shipping as soon as it had something to working with a few design partners until they are thrilled, then doing a proper launch.

Max attributes Legora's pace to Stockholm's talent pool and to his co-founder Sig, who codes about 14 hours a day and climbs for an hour in the morning seven days a week. Legora is upfront in recruiting that it is not a 9-to-5 job; the company requires one or two weeks of onboarding in Stockholm and has been diligent about no remote work because momentum breeds momentum. Max says Legora recruits people who love winning and hate losing, and the losses hurt badly, though the past three or four months have had fewer losses as the company went from about 10 people to 100. It was day-one enterprise: its first client and design partner was Mannheimer Swartling, the largest law firm in the Nordics, and Legora spent half its initial angel funding on SOC and ISO certifications while Max took no salary for six months.

Max calls MCP both underhyped and overhyped. It is underhyped because it makes it possible to serve a universal app into many other capabilities, but it is overhyped because everyone is talking about it and trying to implement it while it has not yet moved into full production. He says authentication and other pieces still need to be gotten right.

The biggest surprise has been how people start using the product. When Legora released its drafting feature, an attorney came in and simply asked it to "write me an SPA," which Max thought was obviously not going to work. Expectation management is a recurring challenge: some associates are savvy, with templates, prompt libraries, and workflows, while others expect the world. That is why Legora runs trainings, onboardings, and office meetings with large firms, and Max says he has thought a lot about how to productize onboarding.

Max changed his mind about the interface. Legora's first version in summer 2023 was completely button- and workflow-based, with a library of nine differently colored buttons for tasks like summarizing, checking a document against policy, or searching a database. Max pushed for chat because he thought it would be a better interface to do more and that functions could be added as tools the chat could use. Moving to chat meant deleting 95% of the source code after Legora was accepted into YC. Outside law, Max is excited about clinical research organizations in pharma, where a lot of manual work and structured workflows could be disrupted by a fully AI CRO, and about voice and audio transcripts for depositions.

Max says managing partners are thinking hard about how to train new associates for the lawyer of the future. Previously firms might have hired underconfident overachievers who were good at following instructions and working step by step. Now they will need entrepreneurial, creative people who challenge existing ways of working, because partners will sit atop pyramids and many processes will be augmented with AI within six to twelve months. He expects every lawyer to become a manager of AI agents from day one, and that individuals, not just innovation or IT departments, will need to augment their own work with AI.