YC Winter 2026 Demo Day reveals 190 startups chasing AI

YC Winter 2026 Demo Day just wrapped with nearly 190 companies pitching, and the batch looks like every other accelerator class right now: AI everything. Y Combinator ran the event on Tuesday with founders presenting across law, healthcare, transportation, and a dozen other verticals. But the common thread? Nearly every pitch included the phrase “powered by AI.”

YC posted pitch videos about 20 minutes after each presentation at YC Winter 2026 Demo Day instead of running a livestream. That format meant watching selectively. I scanned all 190 companies, watched the ones that sounded interesting, and here are 16 that actually stood out.

What YC Winter 2026 Demo Day Actually Showed

Here’s what stood out from YC Winter 2026 Demo Day after watching dozens of pitches. Most are AI tools automating work that used to require humans. Some are legitimately clever. Others feel like solutions searching for problems.

ARC Prize Foundation creates benchmarks to measure progress toward AGI. It’s a nonprofit, which is unusual for YC. But OpenAI, Anthropic, and Google already use their benchmarks. The foundation runs competitions and awards research grants to push open source AGI research forward. Whether we’re close to AGI or not, someone needs to track it.

Asimov collects human movement data to train humanoid robots. People worldwide submit videos of themselves performing tasks, and Asimov turns those into datasets. The idea is to teach robots how humans actually move instead of making them lumber around like, well, robots. Humanoids are years from being useful outside warehouses, but the data collection approach makes sense.

Avoice automates non-design work for architecture firms. Reviewing specifications, drawings, contracts, proposals. The tedious stuff architects hate. The founders said the architecture market is underserved, which tracks. Not many B2B SaaS companies target architects specifically.

Button Computer is a wearable AI device. Two former Apple employees built it. The device connects to email, Slack, Salesforce, and you control it via voice. Everyone’s racing to build wearable AI while waiting to see what OpenAI ships after acquiring Johnny Ive’s company. Button is tiny. Whether anyone actually wants to wear it is the question.

CodeWisp lets anyone build games using AI. Tell the AI what game you want, and it builds it. That’s the pitch. Vibe coding for games. It’ll be interesting to see if this creates a new wave of indie game developers or just floods app stores with more low-quality games.

The Security and Defense Plays

Crosslayer Labs detects website spoofs. Agentic AI tools make it easier to clone websites for scams. Crosslayer monitors customer sites to catch spoofs before they do damage. Security startups solving problems created by AI tools. That’s the cycle now.

Milliray uses radar to track small drones. Defense tech is hot right now, and this targets a real problem. People in the field struggle to distinguish tiny drones from birds. Milliray’s sensors identify what’s actually a drone. Given geopolitics, demand for this tech exists.

Lexius embeds AI into existing security camera systems. The cameras detect theft or falls automatically instead of requiring someone to review footage later. It targets businesses with cameras that lack intelligence. Useful if it works. Most security camera footage gets reviewed only after something bad already happened.

Startups Solving Unsexy Problems

Librar Labs built an AI-powered library management system. Libraries. Not exactly a hot startup vertical. But the founder said there’s almost no competition for modern tools in this space. They’re starting with schools for inventory and cataloguing. Sometimes the best markets are the ones everyone ignores.

Doomersion teaches languages while you doomscroll. You watch short videos like TikTok, but they’re in the language you’re learning. It gamifies the one thing people already do too much of. If you’re going to waste time scrolling, might as well learn French.

Avoice automates administrative tasks for architects. Reviewing specs, contracts, drawings. The work architects went to school specifically to avoid doing. The market is small but underserved. Not every startup needs to go after enterprises with $100M ARR potential.

The Enterprise AI Tools

MouseCat uses AI to investigate fraud. It pulls data from Databricks or Snowflake, analyzes consumer activity, flags suspicious patterns, and recommends actions. Fighting AI-powered fraud with AI-powered detection. The arms race continues.

Sonarly helps software fix its own production issues. It connects to monitoring systems, cuts alert noise, identifies root causes, and either fixes problems automatically or suggests next steps for engineers. Code review tools are everywhere now. But once code hits production, fewer startups are tackling that workflow.

ShoFo bills itself as the world’s video library. It’s a custom video index helping AI labs find diverse datasets efficiently. Search and organization for training data. Not consumer-facing, but useful infrastructure if AI companies keep needing better datasets.

The Wild Cards

Opalite Health uses AI to translate for healthcare providers and non-English-speaking patients. Medical translation tools already exist, but language barriers in healthcare can be life or death. Multiple startups are tackling this. The question is execution and integration with existing EMR systems.

Sequence Markets lets people trade crypto and prediction markets on one platform. Less fragmentation when executing trades. It’s consolidation, which users generally want. Whether traders actually need this depends on how much they’re currently jumping between platforms.

Terranox AI uses AI to find uranium deposits in North America. Uranium powers nuclear energy. Nuclear energy will power data centers. Data centers run AI models. The circle of AI life. The founders didn’t explain how they plan to safely excavate toxic uranium, but that’s an execution detail for later.

What This Batch Actually Reveals

The YC Winter 2026 Demo Day format made it harder to gauge founder energy. Watching pre-recorded pitches posted 20 minutes after the live event feels distant. You lose the room’s reaction, the investor questions, the vibe.

Most YC Winter 2026 Demo Day companies are selling to other startups or enterprises. Very few consumer plays. The batch is heavy on vertical AI tools automating specific workflows. Security, fraud detection, library management, architecture. Unsexy problems with clear customers.

The YC Winter 2026 Demo Day batch is big on automation. That tracks with where the market is. If AI can automate a tedious workflow, someone’s building a startup around it. The question is always the same: can they execute? Building the tool is step one. Getting customers to actually use it and pay for it is where most of these companies will succeed or die.

Nearly 190 pitches is overwhelming. YC keeps growing batch sizes, which dilutes the brand’s selectivity signal. When everyone’s in YC, being in YC means less. But 16 of these startups could actually build something. That’s the bet accelerators make: spray enough capital and advice, and a few will break out.