We Backed 16 Companies from YC’s Winter 2026 Batch. Here’s the Full Picture.
A founder described their product to me recently in a way I keep coming back to.
“It handles the part everyone hates.”
That was it. No grand theory. No speech about reinventing an industry. Just a clear description of a painful, recurring task inside a real workflow, and a claim that software could now take it over reliably enough that a company would pay for it.
That line probably captures this YC batch better than any market map could. Because the energy in Winter 2026 is not about demonstrating what AI can do. It is about deploying AI to do something specific, something expensive, something that a person was doing before and now does not need to.
That shift matters. Early waves of any technology boom tend to produce demos. Later waves produce tools that fit into budgets. This batch has more of the second feeling.
We backed 16 companies this batch across AI infrastructure, developer tools, fintech, robotics, compliance, and enterprise automation. Our standard check in each. Here is who they are, what they are building, and why we said yes.
Agentic Fabriq is building the identity and permissions layer for AI agents, what they describe as Okta for agents. As agents enter enterprise workflows, the question of who they are acting on behalf of and what they are allowed to access becomes a real security problem. Agentic Fabriq sits between agents, users, tools, and data to handle that. Founders Paulina and Matthew are MIT researchers who dropped out to build this. The underlying thesis is that every major computing shift has forced identity infrastructure to evolve, and the agentic era is no different.
Ashr builds testing and evaluation infrastructure for AI agents, specifically by simulating real user behavior inside production-like environments before anything ships. Most teams currently find agent failures in production, which is a bad way to find them. Ashr is the earlier intervention. The founders spent time at Berkeley and have direct backgrounds in AI-native workflow development. The problem is timely and only gets more urgent as agent deployments scale.
Asimov is building the data infrastructure for humanoid robotics, collecting large-scale human movement data from real environments at scale, homes, workplaces, everyday settings, rather than controlled factory settings that lack diversity. Language models trained on trillions of words. Image models trained on billions of images. Robots are starting from scratch, and the bottleneck is not hardware. It is data. The founders bring Scale AI and U.S. Air Force data pipeline experience, and the largest robotics labs in the world are already customers.
Captain is taking a different architectural approach to enterprise knowledge retrieval, distributing queries across multiple LLMs in parallel with what they call an effectively infinite context window and then reducing results to a single output. The claim is that standard retrieval pipelines top out around 78% accuracy and Captain reaches 95% or higher. Founders Lewis and Edgar have deep backgrounds in production RAG systems and hallucination reduction, and spent a summer talking to engineers at Snowflake and Databricks who kept telling them the same thing: there is no good, scalable unstructured data search. Garry Tan is the primary partner.
Chamber puts AI infrastructure on autopilot for enterprises. The simple version is that 30 to 60 percent of reserved GPU capacity sits idle across organizations because of siloed allocations and poor visibility. Chamber’s platform continuously monitors clusters, reallocates resources in real time, detects unhealthy nodes, and lets teams run roughly 50% more workloads on the same hardware. The founding team spent years at Amazon building infrastructure optimization at scale. They have done this before at a much larger size. Brad Flora is the primary partner.
EigenPal automates document processing workflows with AI, the kind of back-office work that still requires humans to read messy scans, handwritten forms, third-party documents, and type the contents into other systems. KYC documents. Invoices. Loan applications. Claims. The founders built an eval-first platform, meaning teams can test a workflow against their historical documents and see exactly what accuracy they get before deploying it. They are already working with two large European banks on loan automations. Jared Friedman is the primary partner.
Lexius layers intelligence onto security cameras that already exist. Most buildings have cameras. Almost none of them have real-time intelligence. Replacing the infrastructure costs $50,000 to $100,000 per site, which is why most organizations never do it. Lexius runs on top of whatever is already installed and delivers live alerts for theft and liability events, cross-camera search, automatic case building, and repeat offender tracking. Setup takes minutes. They are already trusted by 7-Eleven, Erewhon, and Prada. Co-founder Liam Webster co-authored one of the largest open-source computer vision frameworks. Brad Flora is the primary partner.
Menza is an always-on AI data team for consumer brands, connecting to Shopify, Klaviyo, ad platforms, and 650 other sources to surface insights automatically rather than waiting for someone to ask. The pitch is not that it answers questions. It is that it finds the things you did not know to look for, a stockout leaking 10% of monthly revenue, a single email flow change that caused a 34% performance drop, a misattribution hiding $18,000 in a GA4 setup. The retention number says more than anything else here: 100% since launch, with MRR growing 234% since January. Co-founder Qasim previously built ShortlyAI, one of the first commercial LLM applications, which was acquired by Jasper. Co-founder Mariam is a Forbes 30 Under 30 alum who came from the Goldman Sachs trading floor.
Payna helps regulated companies get licensed faster and stay licensed across all 50 states, automating the filing, renewal, and compliance monitoring that currently requires either expensive law firms or significant internal headcount. The founders hit this wall themselves when running a crypto lending company and spent months navigating it manually. Since going live a few weeks ago, they have already signed six figures in annual contract value. The pain is structural, persistent, and directly tied to revenue. Brad Flora is the primary partner.
Polymorph is building living user profiles for consumer and self-serve apps, running inference over behavioral signals to send personalized messages at the right time and channel and then attributing conversions back to what actually worked. Most growth tooling still thinks in static segments and brittle automations. Polymorph’s bet is that software should adapt at the individual user level. The founding team includes David, who led growth experiments at Meta Ads with $400M+ revenue impact, and Andrew, who built Scale AI’s production ML inference infrastructure. Brad Flora is the primary partner.
Protent builds real-time video intelligence for public safety. The problem they describe is specific: one operator is typically responsible for hundreds of live camera feeds, and fewer than one in 200 incidents gets detected in real time. Protent monitors every feed simultaneously, surfaces the most critical streams automatically, and lets teams query all active footage in natural language. They are already working with police departments in Atlanta, Chicago, and St. Louis metro areas. Co-founder Srihan deployed RL-optimized video intelligence at Lockheed Martin. Jon Xu is the primary partner.
Remy AI is building flexible, dexterous robots for warehouses, specifically for the 80% of facilities that have never been able to justify the cost or operational complexity of traditional automation. Their system slots into existing workstations, learns new tasks with significantly less data than standard approaches, and handles the kind of variable, physical work that rigid legacy systems cannot. Co-founder Ben Kaye holds an Oxford PhD in computer vision and 3D reconstruction and built safety-critical systems at OrganOx. Co-founder Oscar comes from supply chain advisory work at BCG. Brad Flora is the primary partner.
RoboDock is building autonomous depots for autonomous and electric vehicle fleets, automating the charging, vehicle checks, and related depot operations that are still entirely manual today even as the vehicles themselves grow increasingly capable. The product retrofits into existing infrastructure without layout changes. Founders Zinny Weli and Celine Wang bring backgrounds that map precisely to the problem: Zinny led autonomous drone charging at Zipline and built the charging system for Amazon’s home robot; Celine was a senior mechatronics engineer at Plus, building systems for autonomous semi-trucks. Gustaf Alstromer is the primary partner.
RunAnywhere provides the SDK and control plane for deploying on-device AI at scale, handling the parts that make shipping edge inference genuinely painful today: model delivery and versioning, lifecycle management, multi-engine abstraction across iOS, Android, React Native, and Flutter, hybrid routing between device and cloud, and observability across fleets. They are already live and open source with more than 10,000 GitHub stars. On-device inference is not a niche: privacy requirements, latency expectations, and offline use cases are all pushing more compute away from the cloud. The infrastructure to manage that shift is still early. Diana Hu is the primary partner.
sitefire is building a marketing suite for what they call the agentic web, specifically for improving how brands appear in AI-driven discovery surfaces like ChatGPT and Gemini. Unlike most tools in this space that stop at analytics, sitefire takes action: it identifies which content is driving AI visibility, recommends and produces AI-optimized blog posts, and delivers them directly to your CMS. They are already working with BMW, Xtrackers, and DWS. Co-founder Jochen brings deep RL and optimization work from Stanford and TU Munich. Distribution is changing, and there will be a category of tooling built around visibility in AI interfaces the way there was once a category built around search. Jon Xu is the primary partner.
SpotPay is a global neobank that gives users one account to receive money from abroad, pay locally, spend anywhere with a SpotPay card, and save on their balance, all built on stablecoin rails. The framing that got my attention was this: most remittance innovation stops at the sender. The receiver is still stuck with legacy ways of collecting and spending what they receive. SpotPay is building for that side. Live in 40+ countries. Total payment volume growing massively week over week. Co-founder Zsika grew up between the Caribbean and Latin America and experienced this problem firsthand. Co-founder Thomas was engineer number four at Brex. Tyler Bosmeny is the primary partner.
Looking across all 16 we invested in so far (we may invest in one or two more), a few things stand out.
The density of AI infrastructure plays is striking, not applications, but the layer underneath them. Identity for agents. Testing for agents. GPU orchestration. On-device inference management. These are the foundational layer of a more autonomous computing era, and if agents are going to work inside enterprises at any real scale, all of this plumbing has to exist first.
Robotics showed up more seriously this batch than in recent ones. That feels right. The convergence of better simulation environments, more capable foundation models, and cheaper compute has created a window where companies can go after physical-world automation problems that were out of reach five years ago.
And several of the fintech plays share something in common: founders who hit the problem themselves before they built the solution. That texture matters. People who have been close to the failure mode build differently than people who identified it from the outside.
That line I mentioned at the top, about handling the part everyone hates? The founders who understand that best are rarely the ones who read about the problem. They are the ones who lived inside it long enough to know exactly where it breaks.
That is what we are betting on.
The fund we invested from in this batch is still open. If you want to get into a marked-up portfolio before we close it this quarter, view the deck here and get in touch.