A five-month-old Swedish startup is asking investors to believe that companies are sitting on valuable patents they barely understand — and Andreessen Horowitz just led a $10.5 million seed round to back that thesis.
Stilta, founded by Oskar Block, Tobias Estreen, Petrus Werner, and Oscar Adamsson, announced the round with backing from Y Combinator and operators from companies including OpenAI, Legora, and Lovable, according to TechCrunch. The sharper read is not just “AI for lawyers.” Stilta is betting that patent analysis has been too slow and expensive for companies to know what they already own.
That makes the startup less a traditional patent-management tool and more an attempt to turn dormant intellectual property into searchable, litigable, monetizable business intelligence.
Stilta’s bet: corporate patent archives may be the next AI gold mine
Block’s origin story is unusually specific. He told TechCrunch he first built machine-learning models for sports betting at 18, later worked on AI integration strategies in consulting, then saw the patent process up close at an autonomous trucking company.
The company’s founding spark came from a dinner conversation with Estreen and Estreen’s father, a patent attorney. Block recalled the attorney describing his days as:
“Reading the same kind of documents, the same way he had for 30 years,”
That line matters because Stilta is not pitching AI as a replacement for the legal decision-maker. It is targeting the pre-decision grind: research, comparison, prior art review, claim mapping, filing history, and court history.
Block’s core argument is that many companies hold patents they have “never enforced, never licensed, never even analyzed properly because the cost of doing so was prohibitive.” If that is true at scale, Stilta is selling access to value that already exists on corporate balance sheets but has remained functionally invisible.
MLXIO analysis: The important question is whether Stilta has found a real enterprise blind spot or is simply surfing the broader rush into AI-native knowledge work. The funding suggests top-tier investors see more than a narrow legal tool. They see a workflow where expensive human analysis can be compressed into software-assisted review.
The seed round signals more than capital
The funding figure comes with a small reporting wrinkle. TechCrunch’s description references a $10 million seed round, while the fuller reported figure is $10.5 million. The source material does not clarify whether that difference reflects rounding, added commitments, or reporting variation.
Either way, the round is large enough to send a signal. Andreessen Horowitz led it. Y Combinator participated. Operators from AI and legal-tech-adjacent companies joined.
That investor mix gives Stilta three advantages beyond cash:
- Credibility: Enterprise legal teams are cautious buyers. A16z and YC backing helps Stilta get taken seriously earlier than most seed-stage vendors.
- Hiring pull: The company is moving from four founders to its first hires, with engineering and go-to-market roles planned in Stockholm and a New York office planned by year-end, according to related source material.
- Network access: Investors and operators tied to AI-native companies can help Stilta refine the product around agentic workflows rather than bolt AI onto old legal software patterns.
The agent angle also connects to a broader MLXIO thread: vertical AI systems are moving from general chat into task-specific execution, a shift we have tracked in Catena Labs Raises $30M to Build Banks for AI Agents and 72% Fara1.5 AI Crushes OpenAI and Google on Web Tasks.
The forgotten-patent problem is a cost problem first
Stilta’s thesis starts with a simple operational failure: companies may create or acquire patents, then lack the economical means to evaluate them continuously.
The source material does not provide a full taxonomy of how patents become neglected. But the pattern is easy to infer from Block’s framing: if analysis is expensive, only the most urgent patents get deep attention. Everything else waits until litigation, licensing talks, or a product conflict forces review.
Stilta is built for that backlog. Users enter a patent number and relevant content. A network of AI agents searches for patents that might conflict with the claim, flags similar IP, and pulls filing and court history.
Block described the system this way:
“They reason in parallel and converge the way a room full of specialists would, but at a scale no human team can match.”
He also stressed that the lawyer remains in the “driver’s seat,” guiding the analysis rather than handing over judgment. That distinction is critical. Patent work is not just search. It is adversarial interpretation backed by evidence.
Stilta is trying to move from search results to litigation-grade work product
The most important product claim is not speed. It is citation-backed output.
Block told TechCrunch:
“The output is litigation-grade: a report and claim charts with pinpoint citations to every piece of evidence.”
Related source material adds more detail. Stilta says its agents can search across 180 million patents, 250 million scientific publications, and more than a trillion archived web pages through an integration with the Internet Archive’s Wayback Machine. The system also pulls prosecution history from the USPTO.
In one demonstrated invalidity analysis against a wireless-networking patent, the platform surfaced 868 prior art references after roughly half an hour and mapped them against claim limitations in a color-coded claim chart. Stilta also says an internal benchmark showed roughly three times the recall of general-purpose models like ChatGPT, Claude, and Perplexity on the same invalidity task.
Those are company-provided claims, not independent audit results. But they show the bar Stilta is setting: not a chatbot that summarizes patents, but an evidence engine for invalidity, infringement, and freedom-to-operate workflows.
| Workflow | Stilta’s stated role | Human role |
|---|---|---|
| Invalidity analysis | Find prior art and map references to claim limitations | Assess legal strength and case strategy |
| Infringement analysis | Compare claims against relevant products or materials | Decide whether to assert or defend |
| Freedom-to-operate | Surface potential conflicts before commercialization | Interpret risk and business response |
| Patent history review | Pull filing, prosecution, and court history | Evaluate legal significance |
Buyers will not all judge Stilta by the same standard
Corporate IP teams may care most about latent value. Related source material says Stilta’s early customers include Roche, Alfa Laval, and Maersk, and that roughly two-thirds of its customers are in-house IP teams across pharma, industrials, and high-tech.
Law firms will judge the product differently. For litigators, speed helps only if the output is defensible. Claim charts, pinpoint citations, source PDFs, and auditability matter more than polished summaries.
Founders and product leaders may see another use case: understanding whether a company’s inventions are actually defensible before fundraising, acquisition talks, or product expansion. But Stilta has not disclosed revenue, pricing, or customer counts, so it is too early to measure how broad adoption really is.
MLXIO analysis: The biggest buyer skepticism will be practical. Rediscovering patents is not the same as creating value from them. Licensing can be slow. Litigation can be expensive. Portfolio strategy can stall inside legal, finance, and business units unless someone owns the commercial follow-through.
Competitors are already circling the same workflows
Stilta is not alone. TechCrunch names Solve Intelligence and DeepIP as companies in the space. Related source material adds Patlytics, which raised a $40 million Series B in April, and notes that Solve Intelligence raised its own $40 million Series B in December 2025 and launched Charts, a product aimed at invalidity, infringement, and freedom-to-operate workflows.
That context sharpens Stilta’s positioning. Block argues that many patent-AI tools have focused on prosecution, while Stilta starts with litigation-grade analysis. His claim is that beginning with the highest-accuracy workflow makes lower-stakes IP work easier later.
Andreessen general partner David Haber framed it even more aggressively:
“Patent litigation runs on labor-intensive workflows that haven’t meaningfully changed in decades. Stilta automates them and, in doing so, becomes the system of record for how enterprises protect and monetize their most valuable intangible assets.”
That “system of record” language is ambitious. It suggests Stilta does not want to remain a point tool for patent searches. It wants to become the operating layer for IP decisions.
The next proof point is whether dormant patents change boardroom decisions
Stilta’s near-term test is not whether its agents can find more documents. It is whether customers use those findings to act.
Evidence that would support the thesis includes recovered licensing opportunities, faster invalidity work, stronger litigation preparation, clearer freedom-to-operate decisions, or portfolio reviews that lead companies to keep, sell, enforce, or abandon patents with more confidence.
Evidence that would weaken it is equally clear: if outputs require too much attorney rework, if enterprise buyers treat the product as a research toy, or if “forgotten patents” rarely translate into commercial decisions, Stilta’s value narrows.
For now, the company has capital, named enterprise customers, top-tier backers, and a sharp wedge into one of the most document-heavy corners of legal work. The unresolved question is whether AI can turn patent archives from static records into live strategic assets.
The Bottom Line
- Andreessen Horowitz and Y Combinator backing signals investor confidence in AI-driven legal infrastructure.
- Stilta is targeting a costly bottleneck that may prevent companies from understanding the value of their own patents.
- If the model works, dormant intellectual property could become a more active source of licensing, litigation, and business strategy.
