Legora Reached $100 Million ARR by Freezing Sales to Rebuild

Max Junestrand says Legora’s rapid growth depended in part on recognizing when selling faster would damage the company. From general availability in October 2024 to the end of the most recent quarter, he says, Legora grew from $1 million to $100 million in annual recurring revenue, while expanding from three engineers in Sweden to more than 750 people globally. The acceleration followed a six-month freeze on sales.

An ICONIQ chart shown during the presentation plots Legora’s stated 18-month route from $1 million to $100 million ARR alongside Wiz, Harvey, Slack, Twilio, Procore, Anaplan, and ServiceNow. It places Legora at the far left of the comparison: about 1.5 years from $1 million to $100 million in ARR.

The decision to stop selling came after an early funding surge. Benchmark invested $9.51 million, Redpoint pre-led the Series A three weeks later, and Legora had about $35 million in the bank with a team of 10. For one month, Junestrand says, the company earned more in interest on its cash than it earned from customers. “That is not a very good sign when you have become a bank.”

His stated rationale for freezing sales was product readiness in a market where a bad first impression could be costly. “In law, you are not paid when things go right, you are punished when things go wrong,” he says. If the system was slow, unavailable under load, or failed in front of a prospective firm, the team believed it could not simply compensate with a stronger sales process. The company stopped selling and rebuilt before its early reputation hardened.

The freeze exposed a product-process problem. In the earliest Leya days, the team decided what to build through democratic votes across the company. That produced too many features and competing priorities. Ahead of a post-summer launch, it needed to rebuild the platform while keeping it adaptable to changing foundation models and agent-development frameworks.

In October 2024, the 25-person company wrote what it called the Leya product manifesto: a simple document intended to collect what the team had learned and give everyone a common direction. A document shown on stage listed a product vision alongside tabular review, a Word plug-in, an assistant, data integration, legal research and sources, and a product process. At the time, Junestrand says, Leya was doing about $1.3 million in ARR while competitors with products focused on only one of those capabilities were doing roughly 10 times as much.

The founders had to narrow their direction without narrowing their ambition. Junestrand says the refocus gave Legora enough product momentum to compete more effectively and make the move to the United States for larger customers. It also changed what the company could credibly sell: not an accumulating list of features, but a platform based on a more coherent view of legal work.

Legora describes itself as an “agentic operating system for lawyers,” intended to handle complex legal work from start to finish. Junestrand says more than 3% of the world’s lawyers are active users. The customer comments he displayed were about recovered capacity rather than novelty: one lawyer credited Legora with making work compatible with golf travel; another said the distance from having an idea to executing it had never been shorter; a third said that if Legora disappeared, he would return to coaching high-school basketball.

Legora’s origin includes legal expertise, but not in the later three-founder team that went through YC. In 2020, before the current wave of generative AI, a lawyer, physicist, engineer, and psychologist started a company called Judilica. They had noticed that law students spent internships summarizing court cases and began exploring what early models such as BERT could do for the field.

Junestrand joined later. He met August and Sigge, his eventual co-founders, at a volleyball game in the Swedish archipelago. Their GPT-3.5 demonstration could explain what a stock-option agreement meant. He initially agreed to “help out a bit,” then dropped out of college and did not complete his master’s thesis because, in his view, GPT-3.5 had made the opportunity cost of not building too large.

The later three-founder team did not include lawyers. Its first method of acquiring domain knowledge was direct exposure: cold emails to lawyers whose contact details appeared on law-firm websites, LinkedIn messages, and invitations to lunch in exchange for the lawyer’s hourly fee. Junestrand says many lawyers met with them, many did not charge them for the time, and some paid for lunch themselves.

That work led them to Mannheimer Swartling, which Junestrand describes as one of the largest law firms in the Nordics. Two years earlier, the firm’s managing partner had said on television that AI was more “artificial” than “intelligent.” Earlier legal-AI products, Junestrand says, had required substantial training and many examples while producing limited results. GPT-3.5 changed what could be built, but it did not remove the need to persuade lawyers that a new system could be useful in their work.

The team moved into the firm’s offices to work close to its lawyers. Their room had no real windows, the air conditioning shut off at about 5 p.m., and an engineer would wave the door around at 6 p.m. to get more oxygen into the room. The setup was uncomfortable, but it put the founders close to actual legal workflows.

That learning was visibly absent in Judilica’s first YC interview. The company applied in May 2023 with a promise to let users access and query legal documents with LLMs. The application was not selected for YC’s Summer 2023 batch.

When a YC partner asked what type of lawyers the company served, Junestrand recalls that the founders responded, “What do you mean? Are there different type of lawyers?” He remembers Tom Blomfield laughing and the team realizing the interview had gone badly.

Two months later, the company reapplied under the name Leya with a more specific description: “Gen AI platform for time-consuming legal tasks.” It was accepted into YC’s Winter 2024 batch. Junestrand treats the change not simply as a persistence story, but as evidence that the team had done the work to understand the market more precisely.

The same distinction shapes his answer to whether founders need domain expertise. A Swedish VC had been interested in leading Leya’s pre-seed but declined because the team had no lawyers. Junestrand says the investor later came to view Legora as evidence that prior expertise is not always decisive. In areas such as quantum computing or fusion, he says, specialized expertise may be necessary. Legal technology was learnable for this team because they were willing to spend as much time as possible with lawyers.

He does not present this as ignorance disguised as confidence. His argument is that founders can enter a market without inherited expertise if they are willing to learn it quickly and directly. Legora’s founders picked a space—legal work and AI—before they had a settled answer to every product question, then learned by moving in that direction with customers.

The conventional view in 2022 and 2023, according to Max Junestrand, was that a vertical AI company needed to fine-tune a model. He recalls Bloomberg spending “millions and millions and millions” on a legal model. Legora took a different view, partly because it did not have the money to pursue that approach and partly because its founders believed frontier models would keep improving.

That did not mean waiting for a finished technology. The company’s task, as Junestrand describes it, was to deliver whatever value models could generate to a particular market under that market’s current constraints. The first generation of ChatGPT was unusable in law, he says, both because it was not sufficiently capable and because it did not support private conversations or European data hosting. His first sales pitch was effectively: Legora was like ChatGPT, but compliant in Europe.

He did not treat that as permanent differentiation. ChatGPT could eventually address those gaps. It was enough for the moment, however. The operating principle was to build for “the world today and maybe like one step ahead,” rather than for an imagined end state of model capability.

The strategy is to capture the opportunity available now, build strong customer relationships, and let rising model capability increase the platform’s value over time. Junestrand says Legora has built perhaps 1% of the software it will ultimately build, and expects the product to look very different within a few years.

Different legal tasks demand different tradeoffs. Customer-support software may optimize for response speed, latency, and cost per resolved ticket. For complex litigation, by contrast, the cost of LLM usage may be negligible relative to the cost of human legal expertise. In those cases, Junestrand says, customers may prefer the most capable model available and let it work for a long period, even if that produces a large model bill.

Legora sees both preferences. Some customers want the most capable models; others want cheaper open-source options. Banks and large law firms may also have governance restrictions, including reluctance to use Chinese models. The problem is therefore not choosing a single model provider. It is routing work according to intelligence, cost, latency, customer requirements, data arrangements, and reliability.

Legora built that muscle by hiring lawyers into roles that were partly customer-facing and partly evaluative. Those lawyers developed use cases that the company could test against its internal evaluations. Junestrand says Legora released its “Legora Bench” after three years of internal work. Earlier, when only a small number of frontier models were viable, publishing results seemed less useful. A larger range of model options changed that.

He says SpaceX and Grok were among the best-performing models on Legora’s benchmarks, particularly relative to cost, although the company could not initially offer them because they were not included in its data-processing agreement. The larger point is not allegiance to a particular model. A changing model market makes evaluation a continuing operating requirement.

Reliability remains part of that requirement. Legora has substantial contracts with model labs and compute providers because, Junestrand says, its system cannot go down while lawyers are relying on it for client work. The question is not just which model scores best. It is whether the system is available, compliant, economical, and dependable when a customer needs it.

The product direction now extends from reactive to proactive agents. For the first three years, a user gave Legora a prompt or a set of instructions and the system performed that work. Junestrand describes the next step as connecting agents to contextual systems and allowing them to act on triggers. A sales team receiving a contract could cause a Legora agent to begin working on it automatically, escalating to a lawyer where needed or executing the contract otherwise. An agent connected to a data room could begin organizing the materials and producing a due-diligence report without waiting for a prompt.

For Junestrand, that is the route to one lawyer producing the outcomes of a 10-person team. It also creates substantial systems problems: Legora spends millions on OCR and document parsing, he says, and must operate those workflows at much larger scale.

Max Junestrand argues that building a product and building a company are different disciplines. One early hiring mistake was selecting for prestigious resume signals—what he calls “Y intercept”—rather than for a person’s upward trajectory.

Someone may begin with a high level of skill, he says, but struggle in an exponentially growing company if they are not improving quickly. Legora shifted toward people with high growth potential who wanted to work intensely. Its top seller is 23 years old, Junestrand says, has sold more than $10 million of Legora, and started without sales experience after leaving university.

Legora’s stated values are “lean in,” “fight for excellence,” and “grow together,” forming the acronym LFG. Junestrand says the profanity is intentional: it filters for people who understand the kind of environment they are joining. Until the company had about 500 employees, he interviewed every non-engineering candidate. He now interviews directors and above, relying on an established cultural foundation to do more of the selection itself.

A particular challenge was Jantelagen, or the Law of Jante, a Scandinavian social code whose rules tell people not to consider themselves special, smarter, more important, or better than others. Junestrand sees value in its humility. It can support a culture in which the best idea wins and junior people feel able to speak. But he also sees a tension between that norm and trying to build one of the fastest-growing enterprise companies in the world.

Legora has tried to assemble what he calls a “cocktail” of American, European, and Asian culture. The company wants the shared-input and lower-hierarchy benefits he associates with Swedish culture, without allowing modesty to limit ambition. It also wants competition without turning colleagues into individual rivals.

That principle is meant to shape the company’s response to outcomes. Big wins are celebrated together. Losses are mourned together, followed by an immediate plan to reverse them. Junestrand contrasts “solution mode” with “blame mode” and treats the difference as a source of momentum.

The internal term blodsmak, literally “taste of blood,” has become shorthand for that intensity. It originated when Junestrand used a Swedish expression for waking up extremely excited about something in an interview. Its English translation produced a headline saying Legora employees woke up with the taste of blood, inviting jokes about vampires and flossing. The company adopted the phrase anyway, as a marker of sustained pursuit.

Stockholm is one mechanism for transmitting those expectations. Every Legora employee globally completes onboarding there. Junestrand believes the practice gives offices around the world a shared feel. He points to a U.S.-based legal engineer who learned that a major law firm had an unresolved document-drafting problem, flew to Stockholm, spent a week with engineers solving it, and returned to the U.S. to deliver the solution.

Gustaf Alströmer raised a question from founders about whether ambition can be learned. Junestrand’s answer is that peer groups can change the scale of what a person considers normal. He describes himself as relatively lazy in high school and his first two years of college, playing video games and doing little beyond what came easily. After finding a more ambitious friend group, he went from no internships to eight jobs in one year.

That, he says, is one of YC’s important effects: peers reset one another’s reference points. He sees the same dynamic inside Legora. Rather than describe himself as a distant boss, he describes himself as a peer in the group running the company. Ambition is intended to be distributed: a CFO can challenge marketing to think bigger, even if that is not the conventional scope of finance.

Competitors supplied focus in the company’s early years. Other legal-AI companies were larger when Legora began, and Junestrand says the company felt it was chasing. If Legora becomes number one, he believes it will need either another external benchmark or a more disciplined commitment to being better than it was yesterday. Competitors can be useful, he says, provided they do not lead to “side questing.” The aim is to keep everyone focused on the same thing.

At three people, Legora had no formal goals; the founders simply tried to maximize the outcome every day. At 750 people, everyone knows the monthly goals, and planning is unavoidable. Communication is the first thing that really breaks as a company scales, Junestrand says. The challenge is preserving the speed of the early company without allowing a larger organization to disperse across incompatible priorities.

That applies especially to the CEO. Junestrand says he must “requalify for the job as CEO of Legora” every quarter because each stage is effectively a new company with different problems. His executive team must do the same. Running sales at $1 million in ARR is not the same job as running it at $150 million.

The company’s pace makes conventional SaaS experience useful but insufficient, in his view. Junestrand contrasts Legora’s trajectory with the familiar “triple, triple, double, double” growth pattern, followed by 40% annual growth as a strong result. He says growth that might take a company from 200 to 1,300 employees over four years is occurring at Legora in one year. The response is to compress timelines, be ruthless about saying no, and understand where each person can create the greatest advantage.

Max Junestrand began Legora introverted and more interested in coding than customer conversations. Sales, employee departures, and leadership required work he did not naturally prefer. The amount a person learns, he says, is a function of the discomfort they are willing to endure.

He still sees a smaller company’s ability to iterate on customer feedback as its central advantage over an incumbent. Even at Legora’s current scale, a customer may describe a particular problem on a call, Junestrand may put it in the product channel, and the team asks how quickly it can turn the problem around and delight that customer. He acknowledges that this is unusual at near-thousand-person scale.

The company must avoid turning every customer request into a new direction. Junestrand says speed has to remain connected to a product vision rather than become constant zigzagging. But he treats fast response as a capability worth preserving: a company should lean into the advantage it has over slower competitors, while letting people concentrate on the work they are uniquely good at.

Technical depth remains necessary even as AI makes it easier to generate code. Junestrand says producing more code with AI is not an excuse not to be a strong software engineer capable of building systems at scale; Legora is hiring 100 engineers. Its engineering problems include proactive agents, document parsing, OCR, and reliability at a scale where lawyers depend on the platform for client work.

Storytelling is the CEO capability Junestrand says he most underestimated. A founder has to sell the company to themselves first: sustaining the work is difficult if they do not believe the story. Then the story must persuade employees, candidates, investors, and customers. Legora competes with AI labs for talent, he says, and must make a credible case for why working there will be better than working at Anthropic.

Its Jude Law campaign was one visible use of that capability. A marketing agency suggested that films about lawyers would be much shorter if their characters had Legora; the idea eventually became an effort to cast Jude Law as the face of an AI-powered legal company. Junestrand says persistence got the company to a yes, on the condition that Law choose his own scriptwriter and cinematographer. The campaign made Legora more recognizable outside law, but Junestrand does not treat it as the company’s foundation.

What made the company, in his account, were the windowless law-firm room, airport calls, and late-night Slack threads fixing a bug before a Monday presentation. Funding rounds, large customer signatures, and billboards are visible. The work that creates them usually is not.