Origami Blue Unicron
getty
Since 2005, Y Combinator has backed over 5,000 companies that have a combined valuation of over $800 billion. The Fall 2025 batch features around 150 startups spanning a diverse range of industries — yet one clear trend stands out: the dominance of B2B software, representing nearly two-thirds of the cohort — a clear signal that this generation of founders is focused on building infrastructure and tools that power other businesses. Beyond B2B, the batch includes consumer startups, healthcare, industrials, in fintech, real estate, reflecting YC’s ongoing evolution toward practical, revenue-driven innovation. These are founders building the rails for the AI era — not just riding it.
The Builders
Who are they — these ambitious founders disrupting the world of innovation?
Multifactor — Zero-trust authentication, authorization, and auditing for AI agents
AI agents are moving faster than security can keep up — and it’s already costing companies. Multifactor brings zero-trust authentication, authorization, and auditing to AI agents, letting teams securely share any online account with a person or AI just by sending a link. With exploits like the “invitation is all you need” Gemini breach making headlines, it’s clear: agentic software needs enterprise-grade protection. Multifactor keeps your AI from going rogue.
Lightberry — The social brain for robots
Robots are ready. Their software isn’t. Lightberry gives humanoids the ability to see, hear, talk, and move like humans — transforming today’s remote-controlled machines into autonomous teammates. The hardware is fantastic — Unitree, Booster, you name it — but for humanoids to become truly useful, they need natural human-like interaction out of the box.
Metorial — Vercel for MCP
AI agents only create value when they can act on real tools and data, but doing that inside an enterprise demands reliability, security, and scalability. Metorial makes MCP enterprise-ready in minutes — with reliability, security, and scale built in from day one. Their serverless runtime hibernates idle MCP servers and resumes them with sub-second cold starts — without dropping connections or losing state. Startups and enterprises can now scale AI agents globally in days, not quarters.
Unsiloed AI — APIs for unstructured data
More than 80% of enterprise data is trapped in unstructured formats — PDFs, slides, images — costing AI teams 40% of their time in cleanup. Unsiloed AI builds APIs that turn messy, multimodal data into structured formats LLMs and AI agents can understand and act on. As AI workflows explode, they’re building the infrastructure that makes documents as computable as databases.
Hyperspell — Memory for AI agents
Hyperspell is the memory layer for AI agents. It unifies context from Gmail, Slack, Notion, and Drive so agents can recall, reason, and remember across company knowledge. Developers use Hyperspell’s API to give their agents organizational context and persistent memory.
Bear — Marketing for AI Search
Bear is solving the #1 burning problem for marketers: funnels weren’t built for AI Search. As blue links fade and ChatGPT becomes the new Google, brand visibility shifts into AI answers. Bear helps brands get discovered inside AI search results — where decisions actually happen. In the age of AI agents, if you’re not found by them, you’re forgotten.
Dome — Unified API for prediction markets
Prediction markets are booming but fragmented. Dome provides a single API for trading across platforms like Polymarket and Kalshi, connecting data, liquidity, and tooling into one ecosystem. One API, every market — it’s the infrastructure powering the next financial frontier.
Hypercubic — AI to maintain and modernize COBOL/mainframes.
The world still runs on COBOL — the 1960s language behind banks, airlines, and paychecks. Billions of lines of legacy code quietly power modern life, but the engineers who wrote them are retiring, and no one’s replacing them. For decades, these black-box systems were too risky to touch. Now, with AI finally able to reason about complex software — letting Hypercubic modernize the backbone of modern life safely.
Parrot — TikTok for language learning
Duolingo made language learning fun, not fluent. The truth is, humans don’t acquire languages through word drills; we learn through context, sound, and repetition. Millions scroll TikTok for hours — what if that screen time actually made you smarter? Parrot turns short-form video into real language acquisition, using the world’s favorite format to build fluency instead of brain rot.
Boom AI — Your AI growth team for e-commerce brands
Most online stores lose 99% of visitors — not because of bad products, but because no one’s there to guide the customer. Every click, question, and hesitation leaks revenue, especially when 56% of interactions happen after hours. Boom uses modern AI to think like your best salesperson — but one that never sleeps. They personalize every interaction at internet scale, turning lost visitors into loyal buyers and cutting acquisition costs for brands.
Everest — AI support engineers for outsourced IT services (MSPs)
Everest builds AI support engineers for managed IT providers (MSPs). Most MSPs still rely on humans for repetitive IT work like password resets and printer troubleshooting, keeping their margins stuck at 10-20%. With AI now capable of solving complex IT and networking issues end-to-end, Everest enables MSPs to deliver proactive support and run on software margins.
Telemetron — Customer Support Platform Built for Hardware
Hardware is getting smarter — but support tools are still stuck in the SaaS era. Traditional helpdesks can’t fix devices that need real diagnostics, telemetry, and firmware insight. Telemetron brings AI-native infrastructure to hardware support, using LLMs to reason across logs, manuals, and live data like an expert engineer. As EVs, wearables, and robotics take off, they’re building the platform that powers the next generation of intelligent hardware support.
Narrative — Infrastructure for AI video processing
Using language models on video is way harder than it should be. When Narrative built their own AI video editor, the vast majority of their time went to infrastructure — not AI. Most teams aren’t touching video agents yet because the tooling doesn’t exist and LLMs still struggle with video understanding. They are building the infrastructure to make video-native AI agents a reality.
Deeptrace — AI agents for on-call
Deeptrace automates the end-to-end on-call workflow, from alert triage to root cause to resolution. On-call has long been one of engineering’s most painful jobs, and existing tools only add noise. Today’s models can finally interpret logs and traces in real time — making this an execution problem, not a research challenge.
Digipals — Building the Future of Social in the Age of AI
Social media isn’t social anymore — it’s ads and strangers. Digipals reimagines connection with an AI-native group chat that helps you coordinate hangouts and stay connected through real-life context. The future of social isn’t scrolling — it’s showing up.
Aside — Real-time answers for sales calls
Sales teams lose deals when reps freeze under pressure. Aside fixes that with AI that retrieves the right answer live, mid-conversation. No more “I’ll get back to you.” Just confident, high-converting calls.
Nucleo — Automated CT scan analysis for oncology care
Nucleo helps oncologists extract insights from CT scans for tumor characterization and treatment — already working with leading hospitals like Stanford, Cedars-Sinai, and Weill Cornell.
Jarmin — 24/7 Machine Learning Engineer employees
Jarmin.ai is your 24/7 ML engineering employee that you hire. Just talk to Jarmin like any other employee and Jarmin handles your AI/ML work from there. Simply talk to Jarmin like you would to any team member, and it takes over your AI/ML work from there — whether that means owning entire initiatives or executing individual tasks.
Caddy — Control all your work apps with your voice
Caddy eliminates clicks and copy-paste chaos by letting knowledge workers simply talk to their computers. Today’s knowledge workers already live in a voice-first world — they speak on WhatsApp, iMessage, and Discord. It’s time our tools listened too.
Lemma — Continuous learning for AI agents
AI agents fail silently when inputs shift or prompts drift. Lemma detects and fixes failures in production automatically, based on cutting-edge Stanford research. The result: agents that learn, not break, over time.
Clad Labs — The brainrot IDE
Chad IDE turns waiting time into shipping time. It integrates gaming and scrolling into the AI coding loop — subsidized by ads and affiliate revenue — making it the Gen Z-native coding tool for an attention-driven world.
The Stories Behind the Builders
Every founder’s journey is different — and that’s exactly what makes this batch so compelling. These teams come from wildly diverse backgrounds, driven by everything from deeply personal pain points to unexpected moments of insight. Some were pushed into entrepreneurship by lived experience; others stumbled into ideas that refused to let go. Together, they represent the full spectrum of modern innovation — bold, resourceful, and relentlessly curious.
Kurush Dubash and Kunal Roy, founders of Dome, began after losing money trading prediction markets, then built the tools they wished existed.
Parrot’s founders turned a co-founder’s failed attempt to impress a girl while learning Spanish into a breakthrough in language learning.
As Julia Hudea explains, “Erik, our co-founder, started learning Spanish to impress a girl on vacation and accidentally fell into a year-long rabbit hole on how polyglots learn languages – emerging fluent in three. At the time, we were building an AI pharmacy assistant together, but none of us cared deeply about it. He kept bringing up language learning until we finally listened. What stood out to me was that his approach mirrored how I learned English (as my second language), through immersion and context, whereas my experience learning French the “traditional” way felt like torture. We did a deep dive on every language learning tool on the market and realized none of them teach the way humans naturally learn.”
Lightberry’s team grew up on sci-fi robots like R2-D2 and WALL·E and decided to bring them to life. “We want robots to be coworkers and companions, not just tools,” founders Ali Attar and Stephan Koenigstorfer said.
Lemma’s founders, Jerry Zhang and Cole Gawin, have been best friends and co-builders since freshman year, both full-ride USC students obsessed with agent systems. They put it straight: “We’re young, scrappy, and will do whatever it takes to make things work.”
Hardware obsession drove Shivani Patel, Co-founder & CEO at Telemetron, who says the team gets genuinely excited about physical products people can hold in their hands.
Jun Kim, Co-founder & CEO of Aside, met his co-founders in high school — where Chanhee Lee launched a Microsoft-backed nonprofit at 17 to teach thousands of kids to code, and Sanghun Lee built a pattern-lock app at 14 that hit 1M downloads and $100K ARR entirely solo.
Manu Ebert, one of the Hyperspell founders, has been building chatbots since age 13. Then he was a cognitive neuroscience researcher studying how the brain learns and stores memory before spending a decade in machine learning. When agents arrived, he saw they were still missing memory, which is why he started Hyperspell with Conor Brennan-Burke.
For Juan Casian and his co-founders, Sergio Garcia and Jose Toscano, Boom is the culmination of more than a decade spent building side by side — first as friends, then as operators scaling a company from zero to tens of millions in revenue. They’ve weathered cash crises, rebuilt teams from scratch, and redesigned entire systems in markets where every mistake carries a real cost. Technically, they built a full lending and risk platform in-house, powering millions of personalized decisions through complex data pipelines. Operationally, they’ve run a 100-person organization and learned how to balance product, engineering, finance, and growth under pressure. Boom is the natural evolution of those lessons — a deeper understanding of data, personalization, and what it truly takes to help businesses grow.
Across the batch, one truth stands out: these founders operate with relentless drive, move faster than the market expects, and reject the idea that today’s limitations should define tomorrow’s companies.
The YC Effect
Ask any founder in the batch, and they’ll say the same thing: YC doesn’t just accelerate growth — it rewires how founders think, build, and dream. What happens inside those three months fundamentally changes a company’s trajectory.
For some, the shift is one of pace and intensity. “YC compresses months of growth into weeks,” said Aman Mishra, Co-founder & CEO at Unsiloed AI. “The real magic is momentum — you leave with a completely new sense of how fast ‘fast’ can be.” Others point to the ruthless clarity the environment demands. As Jun Kim described it, “The environment forces clarity through competition. Everything moves faster, focus gets sharper, and suddenly everything you do is leveraged 10x.”
That clarity often comes from talking to users nonstop — a foundational YC principle that reshaped entire product roadmaps. Connor Waslo, Co-founder of Caddy, shared how his team entered the batch thinking they were solving sales training, only to discover a deeper universal truth after hundreds of conversations: “People just want to get work done faster.” YC didn’t just validate their idea — it revealed the real problem worth solving.
For other founders, focus itself became the superpower. Yolanda Cao, Co-founder of Everest, described her defining lesson in one word: “Focus.” She added, “YC forces focus on what drives our main KPI, and nothing else.” That discipline is often what separates startups that survive from those that scale.
But the impact goes beyond speed — it changes a founder’s ambition thermostat. Karim (Wen) Rahme, Founder & CEO of Metorial, puts it simply: “You are the average of the five people you spend the most time with. At YC, those five are the most ambitious founders you’ll ever meet.” That immersion creates a new normal where bold ideas feel expected, not exceptional.
Vivek Nair, Co-Founder & CEO of Multifactor, turned down substantial VC offers to join YC instead — a decision he is “extremely glad” they made. As he explains, “YC has exceeded our wildest expectations in many aspects, with distribution being the most notable advantage.”
Janak Sunil, Co-Founder and CEO of Bear, captured the mindset shift shared across the batch: “At the beginning of the batch, we used to think practically and cautiously; now, we think nothing but big.”
Others found that YC stripped away distractions entirely. As Sai Gurrapu and Aayush Naik, co-founders of Hypercubic, explained: “YC helps you form clarity. It strips away the noise of building a startup and forces you to focus on building something people desperately need.” For him, the biggest benefit wasn’t tactical advice — it was the mental shift: “Move and fail fast, talk to users, build the inevitable.” Surrounded by founders reinventing every corner of the world, “the optimism for the future is electric,” he added.
In the end, the YC effect is not just about building faster. It’s about thinking bigger, moving with conviction, and surrounding yourself with people who push the frontier forward — every single week.
The Takeaway
The Fall 2025 batch marks a turning point for YC — and perhaps for the startup world. These founders aren’t just building apps; they’re building infrastructure for the AI economy. From unstructured data to humanoid intelligence, the next decade’s backbone is being written now.
These founders are proving that innovation isn’t about hype cycles — it’s about execution, speed, and relentless ambition. And if this batch is any indicator, the next generation of billion-dollar companies won’t just use AI — they’ll define it.