Patent Lawyers Are Drowning in Prior Art. Stilta Built AI That Actually Reads It.

The Macro: Patent Law Is Stuck in a Reading Comprehension Nightmare

I spend a lot of time looking at AI products that claim to “revolutionize” industries, and most of them are thin wrappers on an API call. But every once in a while something comes along that targets a genuine bottleneck, and patent prosecution is one of those bottlenecks that basically nobody outside the legal world understands.

Here is the situation. A patent attorney needs to determine whether an invention is actually novel. That means reading prior art. Sometimes that means reading hundreds of related patents, technical papers, and prosecution histories before forming an opinion. This is not a vibe check. Every claim has to be mapped to specific language in specific documents, and the whole analysis needs to be defensible if it ever gets challenged.

The traditional way to handle this is to throw human hours at it. Junior associates and patent agents spend days on a single invalidity analysis. The work is meticulous, expensive, and slow. Firms bill for it. Clients pay. Nobody is thrilled about the timeline.

AI should be perfect for this. It is a reading comprehension task at massive scale. But most AI tools in legal have a fatal flaw: they hallucinate. And in patent law, a hallucinated citation is not just embarrassing. It can be malpractice. So the entire legal profession has been watching AI from a distance, arms crossed, waiting for someone to solve the attribution problem.

That is the gap Stilta is trying to fill. Not general-purpose legal AI. Not another chatbot for contracts. Specifically: agentic AI for patent attorneys that provides source-backed, auditable analysis. Every conclusion points back to a document. Every document is verifiable. The AI shows its work.

The Micro: Four Founders, $20K MRR, and Fortune Global 50 Clients

Stilta came out of Y Combinator’s Winter 2026 batch, which immediately puts it in a specific tier of seriousness. The founding team is four people out of Sweden: Oskar Block as CEO, alongside Oscar Adamsson, Tobias Estreen, and Petrus Werner as CTO. The Scandinavian engineering talent pipeline continues to punch above its weight.

What caught my attention is the speed of the traction. They launched in February 2026 and hit over $20K in monthly recurring revenue with clients that include Fortune Global 50 companies and leading IP firms. That is not a toy. That is enterprise software finding product-market fit in a notoriously conservative industry.

The product itself is built around what they call “agentic” analysis. Rather than a single prompt-and-response interaction, Stilta’s system breaks down a patent question into subtasks, works through them systematically, and assembles a comprehensive analysis with source citations throughout. Think of it as the difference between asking someone a question and handing someone a research assignment with specific deliverables.

The competitive landscape here is worth mapping. ClearstoneIP, PatSnap, and IPRally all operate in the patent intelligence space. Lexis+ AI and Westlaw Edge are making moves from the general legal AI direction. But most of these tools are search-first. They help you find relevant documents. Stilta is analysis-first. It reads the documents for you and tells you what they mean for your specific claims, with receipts.

This is a meaningful distinction. Patent attorneys do not need another search engine. They need something that can do the analytical heavy lifting that currently takes a team of humans multiple days. If Stilta delivers on that promise, the ROI math for law firms is straightforward and compelling.

The auditable piece is what separates this from the pack. Every conclusion maps to a source. Every source is checkable. This is table stakes for legal work but surprisingly rare in AI products targeting the space. Most legal AI tools still produce output that a lawyer has to independently verify from scratch, which defeats the purpose.

The Verdict

I think Stilta is one of the more promising vertical AI plays I have seen recently, and here is why: the problem is real, the market has money, and the founders appear to have shipped a product that actual paying customers are using at scale within months of launch.

The risk is concentration. Patent law is a big market, but it is also a specific one. If they nail invalidity analysis and prior art search, the question becomes whether they expand horizontally into other patent workflows or vertically into other areas of IP law. Both paths have merit. Both paths have competitors waiting.

I would also want to know more about their data pipeline. Patent databases are messy. The USPTO, EPO, and WIPO all have different formats, different access models, and different quirks. Building a reliable ingestion layer for global patent data is genuinely hard infrastructure work, and it is the kind of moat that gets wider over time if you do it right.

At $20K MRR and growing, they are past the “interesting experiment” phase. The next milestone is whether they can land a top-10 IP law firm as a named reference customer. That would signal to the rest of the market that this is safe to adopt. Patent attorneys are herd animals when it comes to technology adoption. One big firm moves, and the rest follow within 18 months.

Four people from Sweden building AI for patent lawyers is not the startup pitch you hear at every demo day. That is exactly why I think it has a shot.