When Oskar Block watched patent attorneys at work during his years at McKinsey, one thing stuck with him: the sheer amount of time they spent switching between systems. Research in one tool. Drafting in another. Portfolio tracking somewhere else entirely. And when they finally tried to incorporate AI assistance, the technology spat out polished-looking documents that fell apart under scrutiny.
Twenty to forty hours per utility application. Weeks of back-and-forth. Context lost in translation between platforms. Block figured there had to be a better way.
Now, with Stilta—a four-person startup that emerged from Y Combinator's Winter 2026 batch—he's betting he can prove it.
The company went public in early February with what it calls "Cursor for patent practitioners," borrowing the name from the AI coding assistant that's become ubiquitous among software developers. The promise sounds almost too tidy: agentic AI workflows that slash manual research by up to 60% and collapse drafting cycles from weeks into days. In an industry where precision matters down to individual claim language, that's either the future or a recipe for malpractice, depending on who you ask.
Context That Doesn't Evaporate
At its core, Stilta is making a bet on how AI should actually work in high-stakes legal environments. Rather than the typical one-shot document generation that's become table stakes for generative AI tools, the platform deploys specialized agents for discrete chunks of patent work: prior art searches, portfolio analysis, infringement monitoring, litigation preparation. Practitioners steer these agents in real time, adjusting parameters and following leads as they emerge.
The company describes this as "interactive, argument-led workflows," though that phrase undersells what's perhaps the more crucial feature—traceability. Every output links back to source material. Every assertion can be verified. In legal work, where a single hallucinated citation can blow a deadline or sink a filing, that's not a nice-to-have. It's existential.
Stilta claims it's already working with what it characterizes as "leading firms in Europe and the US," though the company hasn't named customers publicly. Pricing remains undisclosed. The go-to-market strategy leans heavily on demos and design partnerships with established IP practices—a slower burn than typical SaaS land-grabs, but arguably the only path that makes sense when your buyers need to trust you with work product they'll defend before federal examiners.
Consultants Building for Lawyers
All four founders—Block, Oscar Adamsson, Tobias Estreen (Chief Architect), and Petrus Werner (CTO)—come from McKinsey or its AI spinout QuantumBlack. Block previously bootstrapped two startups past $1 million in annual recurring revenue. Werner put in time at AWS before McKinsey. Estreen brings a physics and machine learning pedigree from KTH and KU Leuven.
It's a team built for enterprise sales and complex systems. Not legal practice. Whether that's an asset or a liability depends partly on your view of how much domain expertise matters when building tools for work that demands word-level precision. Sometimes outsiders see solutions insiders miss. Sometimes they build elegant systems that don't survive contact with actual practitioners.
The Swedish entity, Stilta IP AB, incorporated in January 2026 with all four on the board. Gustaf Alströmer is listed as the YC batch partner—a detail that signals the company has at least one partner with consumer product instincts to balance the consulting DNA.
A Crowded Field, Different Angles
Stilta is hardly alone in chasing AI-enabled patent work. The past year has seen something of a cambrian explosion in the space.
IPRally launched its Graph AI-powered search engine late last year. Perplexity dropped a consumer-facing patent search agent in October, because apparently everyone needs to check if their napkin sketch infringes on Samsung's portfolio. DeepIP raised $15 million in Series A funding in March 2025 for AI-assisted drafting. And &AI, another YC company, pulled in $6.5 million last February for litigation-focused agents.
The competitive positioning varies wildly. Some tools—Ambercite, IPRally—focus narrowly on search. Others automate discrete tasks: office action responses (Patent Bots, PowerPatent, Clairvolex) or specification generation (PatentPal). A handful, DeepIP and &AI among them, are pursuing broader collaboration platforms that follow attorneys through entire prosecution workflows.
Stilta's thesis is that practitioners don't want a collection of point solutions. They want a workspace that maintains context across a full matter lifecycle—initial research through filing, prosecution, portfolio management. The Cursor comparison is intentional. Just as developers have learned to loop AI into their coding flow without surrendering the keyboard, patent attorneys should be able to iterate with AI on complex analysis without losing control of the output.
It's a clean theory. Execution is another question.
The Trust Problem
The patent bar remains, to put it mildly, conservative. Recent surveys show practitioners treating AI as a double-edged instrument—efficiency on one side, accuracy and confidentiality risks on the other. Stilta's emphasis on traceable outputs addresses some of that wariness, but proving reliability at scale is a different game than winning design partnerships with early adopters willing to experiment.
There's also the issue of what happens when the AI gets something subtly wrong. Not hallucinating a fake case citation, which is easy to catch. But mischaracterizing prior art in a way that only becomes apparent three months into prosecution, or overlooking a relevant portfolio entry during an infringement analysis. The kind of mistake that wouldn't be obvious until it costs a client real money.
These aren't hypothetical concerns. They're why every AI legal tool has to clear a higher bar than, say, an AI marketing copy generator. Get a blog post wrong and you edit it. Get a patent application wrong and you might lose your license.
What Comes Next
Y Combinator's Winter 2026 Demo Day is set for March 24, which should bring Stilta more visibility and likely a seed round announcement. For now, the company appears to be running on standard YC funding—around $500,000, if you go by batch norms—and concentrating on proof-of-concept deployments with early firms willing to kick the tires.
You can reach the team at [email protected]. They've posted a launch video on YouTube, though the main website requires JavaScript to render, which suggests a team still prioritizing product development over public-facing polish. That tracks for a startup barely six weeks past stealth mode.
The broader question isn't whether AI will eventually reshape patent practice. That much seems inevitable. What matters is which specific workflows break first under the weight of new technology, and which tools practitioners actually trust enough to adopt when their professional reputations are on the line.
Block and his team think they've found an answer. The patent bar, famously slow to change, will have the final word.
