How Legora Built an AI-Native Legal Software Company

A Swedish expression about working until one can “taste blood” became an unintentionally alarming English quote, jokes Legora CEO Max Junestrand. The anecdote opens a discussion with host Jack Altman and Benchmark investor Chetan Puttagunta about building a fast-growing legal AI company and the culture behind it.

Puttagunta invested when there were already competitors in legal AI. Having backed legal software before, he understood something of the market and was struck by Junestrand’s clear view that general-purpose models would improve and could benefit legal work. The team’s determination to learn from lawyers also stood out.

In 2023 and 2024, Junestrand says, some companies focused on training or fine-tuning their own models. Legora instead bet on improving foundation models and on the application work needed to make them useful: privacy, compliance, file uploads, parsing and document handling.

The early team worked inside a law firm in Stockholm, learning how its documents and data fit together. Junestrand also cold-contacted lawyers on LinkedIn, offering to pay for lunch and their time so he could ask about their work. The founding team was technical, and its first lawyer joined nine months into the journey.

Junestrand explains that the project began in 2020 with four original founders and early language models, before he joined. The arrival of more capable large language models changed the opportunity. After a change in the founding team, they committed to the intersection of AI and law while still discovering what the product should be.

Their first lawyer had been a technology leader at a prospective customer and had built an early connection between a language model and a law firm’s document system. He chose to join the startup. The team’s customer access helped it learn directly from the workflows it hoped to improve.

Altman asks whether rapid legal adoption was surprising. Junestrand argues that law firms compete closely: if one firm improves speed, price or service through AI, others have an incentive to follow. He distinguishes that dynamic from in-house legal departments, where a rival bank’s adoption may not create the same pressure.

He also sees unmet demand from a sector that had long lacked effective software for certain tasks. Puttagunta adds that lawyers are sophisticated users who can try general tools themselves; a vertical product must offer enough additional value to justify paying for it alongside models and Microsoft software.

Early product work involved guardrails, citations, retrieval, context limits and choosing among models for different tasks. Junestrand says the product has since moved toward an enterprise platform: giving models tools and an appropriate environment, then providing an interface through which people can use and review the work.

As models improve, the company must keep justifying the difference between a vertical product and a general model. Puttagunta says that shaped a technical organization: engineers and researchers needed to understand model capabilities while accepting that some features might become unnecessary within months.

Junestrand says the founding engineers hired technical peers and built their own agent framework when existing tools were insufficiently customizable. For due diligence, a simple chat interface was a poor fit for reviewing hundreds of documents, so the team built a tabular review product that runs many document questions in parallel.

A small engineering group was given ownership of that product almost as if it were its own company. Junestrand says many people in the engineering, product and design organization had previously founded startups. For a company entering a market after competitors, he saw product quality and reliability as the way to earn a place.

He recalls a period of roughly six months when Legora held back sales because the product was not ready for rapid onboarding at scale. The company had raised its seed round in March 2024 and planned general availability for October; a month before launch, the team and Puttagunta met in Sweden to decide what mattered most.

The conclusion was to focus on three use cases instead of attempting fifteen at once. Junestrand cites tabular extraction and integrations with Word and Outlook as examples: make the product available where lawyers already work, rather than trying to solve every conceivable legal task immediately.

The company, then called Leya, rallied around a short product manifesto. Junestrand recalls revenue rising from about 1.5 to 4 and then 8 in subsequent quarters; he does not specify the unit in this portion of the conversation. The team began building a US presence as confidence in the approach grew.

Junestrand sees another shift ahead: agents, as well as people, are becoming users of the tools Legora has built. He says features such as tabular review and Word editing increasingly need to be usable by an agent acting for a person, as well as directly through the interface.

Altman asks what makes a company AI-native. Junestrand says teams must be willing to discard earlier work as capabilities change. In his view, assumptions inherited from conventional software architecture and established corporate functions deserve to be reexamined when the cost and speed of building change.

He describes a culture that optimizes for the company rather than protecting each function or feature. Executives, he says, must keep growing with the business. He applies the same standard to himself, noting that he entered the role without extensive prior experience and had to adapt as the company expanded.

Junestrand suggests that AI can make a smaller engineering team more productive, although his comparison with a hypothetical larger team is an estimate. He sees the harder questions as deciding what to build and designing an experience that works for a large user base.

Puttagunta describes a short planning horizon: the team responds quickly as models change and invests in evaluation infrastructure. Junestrand says a model improvement can render months of custom scaffolding unnecessary. That requires people to accept from the outset that useful work may later be deleted.

At the same time, building too far ahead of customers is risky. Junestrand says legal organizations need a path to adopt new capabilities and experience success incrementally. Faster model releases have shortened the interval in which Legora can take a capability, turn it into a product and help customers use it.

He argues that, for many applications, the surrounding software and trustworthy human review have become more pressing bottlenecks than waiting for another model release. He still welcomes improvements such as larger context windows, but says the company has substantial work to do integrating existing capabilities into real workflows.

Evaluation was a priority early on. Customers contributed real tasks and criteria for judging results, allowing the team to compare models against legal work rather than only generic benchmarks. Junestrand recalls a Danish firm’s limited-partnership agreement review exercise in which the team’s reported accuracy rose from 60% to 100% by the end of a summer.

That 100% is his account of one evaluation, not proof of perfect performance on all legal documents. His product ambition is to automate progressively more complex tasks end to end, from simpler agreements toward more demanding transactions, while customers face the separate challenge of absorbing rapid change.

Altman asks how urgent scale is in this market. Puttagunta recalls that at launch the team concentrated on getting the product into users’ hands and learning from feedback, rather than assuming it knew enough to produce a precise budget. Early product exposure was necessary to test the underlying thesis.

One difference from a conventional software trial was flexibility. Puttagunta says Legora allowed customers to retain the practices and work they developed during a pilot; Junestrand offered pilots of different lengths. Firms often compared vendors in real use rather than selecting purely from feature lists.

Junestrand says the pilots demonstrated value strongly enough that users wanted continued access. The company also hired lawyers as “legal engineers” who work with customers on adoption. He argues that recurring use and established workflows, rather than technical lock-in alone, make the product sticky.

Puttagunta emphasizes speed in customer interactions as well as in engineering: remove friction when a customer wants to try or buy. He recalls Junestrand flying from San Francisco to New Delhi because a large Indian firm was interested, rather than waiting for a conventional sequence of regional expansion.

A European starting point also forced the team to think across borders. Junestrand says the Swedish legal market alone was too small, so the company expanded to neighboring countries. Its first major Swedish customer introduced him to firms in other markets, and later expansion to the UK and US accelerated growth.

Before opening a US office, Junestrand wanted evidence that Legora could win and serve major American firms from Stockholm. He says Cleary Gottlieb and Goodwin became customers after competitive pilots, with the team traveling frequently to support them. That gave him confidence to establish a local office.

Puttagunta offers a hypothesis: selling across multiple European countries can force legal software teams to support different languages and legal systems early. He remembers Legora’s small team demonstrating support for Sweden and Spain at the first meeting, a sign that it was already designing for more than one market.

The discussion turns to culture. Puttagunta describes a technical, intense team with a desire to win, and says Stockholm became a recruiting hub for people across Europe. He argues that working with fast-changing models requires humility: a feature a team labors over may soon be superseded by a foundation model.

Junestrand describes long hours and dinners served at 8 p.m. as part of Legora’s intense culture. He says the company is open about these expectations when recruiting and wants to compete for the leading position. These are the company’s stated practices and ambitions, not a universal prescription for building a business.

To carry that culture across offices, new hires onboard in Stockholm, including people joining distant offices. The company also sends existing employees to seed new locations. Junestrand and Puttagunta say the goal is to create a common operating cadence rather than letting each office develop entirely independently.

Junestrand returns to the Swedish expression that became a joke inside the company: after a demanding effort, someone can feel as if they taste blood. Translated literally in an interview, it sounded much stranger. He says employees turned the awkward wording into a shared joke.

Asked about the latest financing, Junestrand says earlier rounds after the seed investment were largely preemptive. He recalls negotiating the seed investment with Puttagunta down to exact share counts, balancing ownership and dilution. The story illustrates the detailed bargaining behind what both describe as a successful partnership.

For the Series D, he worked with newly joined CFO David and says the round drew substantial investor demand. The video description states that Legora raised $550 million at a $5.55 billion valuation in March 2026. Junestrand welcomes the investors and says the company is continuing to expand.