AI again took center stage at Y Combinator’s latest Demo Day, a recurring signal of how deeply artificial intelligence has embedded itself across the startup world. Nearly 190 companies—spanning everything from law to logistics to healthcare—made up the Winter ’26 cohort, each jostling for the spotlight and an investor’s quickened pulse.
This year, media didn’t get front row seats or a livestream; instead, YC uploaded pitch clips about twenty minutes after each presentation. I didn’t attempt the impossible feat of watching every single one. Instead, I combed through descriptions of all 190 startups, then streamed those that caught my eye. Out of this crush of innovation, sixteen stood out—companies that managed to break through the noise with ideas either surprisingly original or weirdly obvious in retrospect.
First, the ARC Prize Foundation. Not many nonprofits make it through YC’s gauntlet, but then again, ARC is already powering benchmarks for AI giants like OpenAI, Anthropic, and Google. Their mission: fuel open-source artificial general intelligence research with competitions and grant funding. The importance is obvious—if the world is racing toward AGI (Nvidia’s Jensen Huang claims we’re already there), someone needs to document how close we’re really getting.
Asimov, with a wink to its namesake, is part of the next wave trying to teach humanoid robots to move like humans—grace and all. Instead of relying on synthetic movement, Asimov gathers crowdsourced videos of people performing everyday motions and turns them into datasets for robot learning. It’s an ambitious effort to make humanoids less mechanical, maybe even a step closer to the “Rosey the Robot” future cartoons once promised.
From there, the spotlight swings to Avoice, a startup targeting the oft-neglected architecture sector. Architects are creative by design, but mired in reviewing contracts and blueprints. Avoice’s AI aims to free them from that drudgery, automating the reading and analysis of drawings, specs, and legalese so designers can get back to, well, designing.
Hardware hasn’t been left behind. Two former Apple insiders have teamed up for Button Computer, a compact wearable designed to harness AI for the everyday user. Picture a voice-activated miniature computer that taps into email, workplace apps, even CRM dashboards. As hype continues to grow around wearable AI (thanks, in part, to OpenAI’s headline-making hardware ambitions), Button looks poised to ride the next hardware wave.
On the play side, CodeWisp lets anyone make games just by telling the AI their idea. It’s a simple pitch—describe your game, the AI builds it. As someone who remembers childhood frustration with clunky game makers, this feels like a little piece of magic, turning raw imagination into code.
Cybersecurity is evolving, too. Crosslayer Labs is racing to spot (and thwart) phishing websites, an ever-growing menace as agentic tools make spoofing more accessible to bad actors. Their system monitors for copycats, catching things the naked eye—and even some security teams—might miss.
Meanwhile, Doomersion found a silver lining to the hours spent doomscrolling social media: why not learn a language while marinating in social feeds? Its app slips short, TikTok-style videos in your target language into your scroll, hacking the brain’s incessant craving for a quick hit of content.
Some teams are thinking brick and mortar—literally. Librar Labs is betting on AI-powered library management systems, starting in schools where keeping track of shelves and catalogues is still surprisingly medieval. No tech giants are eyeing this sleepy industry, so perhaps the time for a “next big thing” in libraries is now.
Defense hasn’t escaped the YC treatment. Milliray uses radar to distinguish drones from birds—no small feat as tiny UAVs fill the skies, complicating geopolitical security and day-to-day surveillance.
There’s fraud, too—both fighting it and, cynically, helping prevent the abuse of big data. MouseCat pulls from vast clouds like Databricks or Snowflake, scanning for anomalies and flagging potential scams, offering actionable next steps. As “bad AI” gets smarter, keeping ahead is a non-negotiable.
Other startups, like Opalite Health, are using AI as a kind of medical glue—translating between doctors and patients who don’t share a common language. In a country where life-saving information can be lost in translation, the stakes are sky-high.
The quest for seamlessness is ongoing: Sequence Markets unites disparate assets—crypto, prediction markets—onto a single trading platform. ShoFo aims to be the “world’s video library,” curating clips for AI labs who crave more diverse data, but also helping the rest of us organize the digital noise.
Sonarly works in the background, promising to reduce alert fatigue for engineers while finding the root of software failures before they spiral. In the energy sector, Terranox AI hunts for uranium deposits, betting that nuclear will fuel the looming explosion in data centers needed to power this same AI revolution.
YC’s Winter ’26 batch paints a vivid tableau: the technical and the practical, the high-concept and the nuts-and-bolts, all woven with the thread of AI. Some of these ideas feel inevitable. Others tilt against the ordinary in search of something goofy, profound, or quietly transformative. The real magic, as always, will be in the execution.
