Oskar Block's Stilta raised $10.5M to put AI agents inside patent litigation

Why it matters

Stilta is a test of whether legal AI can move from drafting assistance into evidence-heavy workflows where auditability, security and attorney control decide adoption.

Oskar Block's Stilta raised a $10.5 million seed round led by Andreessen Horowitz after pitching investors on a narrower legal-AI wedge: patent litigation work where every answer has to be traced back to evidence. Business Insider published the pitch deck on July 8, after the Stockholm-born startup announced the round on May 19, as reported by TechCrunch.

The company is young even by seed-stage standards. Business Insider says Stilta was founded in January 2026 by four former McKinsey colleagues or McKinsey-adjacent operators: Block, Oscar Adamsson, Tobias Estreen and Petrus Werner. LawSites previously described the startup as founded in December 2025. The safe read is that Stilta went from formation to a major U.S. venture round in roughly half a year.

That speed is part of the story. Block and Adamsson came out of McKinsey, where they learned the deck-and-enterprise-sale grammar that still matters when selling software into legal departments. Estreen and Werner came from QuantumBlack, McKinsey's AI arm. Y Combinator, which lists Stilta in its Winter 2026 batch, describes the team as four engineers from McKinsey and QuantumBlack who had built and deployed AI systems at large organizations.

Block's founder arc gives Stilta a cleaner origin than the usual legal-AI wrapper. YC says he started a company building machine-learning models for sports betting when he was 18, later bootstrapped another startup to more than $1 million in revenue and more than 25,000 monthly active users, and previously worked at McKinsey, Goldman Sachs and two Swedish unicorns. Adamsson, Stilta's CPO, worked with more than 50 CEOs at McKinsey on Sweden's innovation agenda, according to YC. Estreen previously led agentic-AI work at QuantumBlack and has a physics and machine-learning background from KTH and KU Leuven. Werner, Stilta's CTO, built and deployed AI systems at QuantumBlack and Amazon Web Services.

The wedge is evidence, not drafting

Stilta is entering patent work from the litigation and analysis side, rather than the drafting workflow that has absorbed much of the first wave of legal AI. The product asks a lawyer or IP professional to define a matter objective, such as finding infringement, defending against a lawsuit, running invalidity analysis or assessing a patent portfolio. Stilta then says its agents search patents, scientific literature, file histories, web archives and other technical sources, map evidence to claim elements, and return source-cited outputs that attorneys can review.

Business Insider quotes Block's commercial thesis plainly: "A lot of companies today sit on this huge patent portfolio. They typically just place them in a drawer." His pitch is that companies already own valuable rights, but the cost of searching, mapping and arguing over those rights keeps much of that portfolio inactive.

The distinction matters because patent litigation is hostile terrain for general-purpose AI. A wrong answer is not a bad draft; it can miss prior art, misread a claim limitation, or send a company toward a claim chart it cannot defend. Stilta's site leans into that anxiety. The company says its outputs are source-backed and auditable, that attorneys remain in control, and that customer data is not used for model training. Stilta also says it supports cloud, private-cloud and on-prem deployments, and lists ISO 27001 and GDPR compliance, EU AI Act compliance and SOC 2 Type II in audit.

The company's public product pages describe three core modules: infringement analysis, invalidity analysis and freedom-to-operate work. For infringement, Stilta says its agents map product features, technical documentation and public disclosures against patent claims. For invalidity, Stilta says agents search prior art and return steerable results instead of a static list. For freedom-to-operate, Stilta says the software decomposes a planned product or launch into features, then ranks potentially risky patents with source-backed evidence.

Those claims should be read as company claims. Stilta has not disclosed audited performance data, revenue, pricing or headcount. Its website says a first defensible answer can be produced in about 17 minutes and references a benchmark across 40 PTAB institution decisions, but the benchmark is company-published. The more durable fact is the design choice: Stilta is trying to make citations, audit trails and attorney control the product, not compliance copy pasted onto a chatbot.

The cap table is a distribution bet

Andreessen Horowitz led the seed round. Other investors include Y Combinator and employees or operators from OpenAI, Lovable and Legora, according to Business Insider. Stilta's own site also says Microsoft backs the company, alongside YC and founders and operators from unicorns. The valuation was not disclosed.

Block told Business Insider that Stilta had multiple meetings with Andreessen Horowitz during a "very short" fundraise and that the team prioritized fit with a16z, turning down other term sheets. That is founder positioning, but it is also a practical go-to-market choice. Patent litigation software needs credibility with firms, in-house counsel and technical teams that will be slow to trust a new vendor with privileged work product and unpublished inventions. A16z gives Stilta a U.S. enterprise network before the company has the history that legal buyers usually want.

LawSites described the seed as Stilta's first outside funding.

Business Insider says about 60% of Stilta's revenue comes from law firms and 40% from corporate legal departments, citing Block. It named Roche and Maersk as enterprise customers. LawSites previously reported Roche, Alfa Laval and Maersk as customers and said the company had signed three of the world's five largest IP firms as customers or active pilots. Stilta's own site shows customer or testimonial logos including KUKA, Scale LLP, Alfa Laval, Bergenstrahle, EIP, Loza & Loza, Highland Advisory Group and Kapea.

The revenue split is useful because it shows where Stilta is trying to sit. If law firms are the majority of revenue, Stilta is selling into the people who already do the work. If corporate legal departments become the bigger share, Stilta could shift some patent analysis back inside companies that currently send large portions of that work to outside counsel. Block's argument to Business Insider was that in-house teams could rein in legal bills by keeping more patent work inside.

Patent AI is filling out fast

Stilta is not entering an empty category. Business Insider named Patlytics and DeepIP as competitors. LawSites said Block identified Patlytics and Solve Intelligence's Charts as the closest competitors, while arguing that many patent-AI tools have focused on prosecution and drafting rather than litigation.

The funding pace supports that read. Patlytics announced a $40 million Series B led by SignalFire on April 8 and said total funding stood at about $65 million. DeepIP announced on March 3 that it had raised a $25 million Series B, bringing total funding to $40 million. Fearn, a patent-drafting startup, announced a $5.5 million seed round on June 10 led by Kindred Ventures, with participation from a16z Speedrun.

That gives Stilta's seed a clear context. Patent AI is moving from application drafting into the parts of IP work where the buyer pays for risk reduction: infringement, invalidity, freedom-to-operate and portfolio assessment. Those workflows are expensive because they require technical search, legal reasoning and defensible evidence. They are also attractive software markets because the work is recurring, high stakes and fragmented across outside counsel, search firms, spreadsheets and point tools.

Stilta's deck, as shown by Business Insider, framed the opportunity around a $55 billion litigation market estimate and showed a redacted MRR growth chart from $0 to "$[X]k" in five weeks. The redaction leaves the traction claim mostly unusable from the outside. What can be said is narrower: Stilta convinced one of Silicon Valley's most visible venture firms that a four-person team with McKinsey and QuantumBlack experience could sell AI into one of the least forgiving corners of legal work.

That is the founder bet behind the round. Block is not trying to persuade lawyers that AI can write nicer prose. He is trying to persuade patent professionals that agents can do the document chase, evidence mapping and first-pass analysis without breaking the chain of trust that makes the work billable, defensible and usable in a dispute. The company has capital, early customer names and a category that is getting crowded quickly. The unanswered question is whether Stilta can turn a credible seed story into a system attorneys use when the matter is live, the client is exposed and the citations cannot fail.