YC Has 1,477 AI Startups. That Says Everything.
AI has swallowed YC so completely that startups are no longer just building copilots. They are building agents that hire, call, trade, code, and operate in the real world.
For a while, Y Combinator started to feel strangely normal.
Every batch seemed packed with the same kind of company: B2B SaaS, AI copilots, workflow automation, dev tools, compliance software, sales tools, support tools, finance tools, and yet another “AI teammate” for some painfully specific enterprise department.
Useful? Definitely.
Exciting? Not always.
But now YC startups are getting weird again, and the reason is simple: AI has taken over so completely that founders are no longer just building AI features. They are building companies around AI as workers, customers, callers, traders, operators, coders, and even buyers of human labor.
YC’s own Startup Directory makes the shift obvious. A recent snapshot showed 1,477 startups under “AI”, while another related directory tag listed 842 under “Artificial Intelligence.” The counts can shift as YC updates its directory, but the message is clear: AI is no longer just one category inside YC. It is becoming the default language of the accelerator ecosystem.
YC used to be the place where “software eats the world” felt like a founding religion.
Now it looks more like AI eats YC.
And that is exactly why the startups are getting strange again.
The AI Agent Is Becoming the Customer
One of the clearest examples is RentAHuman, a startup built around a marketplace where AI agents can hire humans to complete real-world tasks.
That sentence sounds like a joke from a dystopian sci-fi sketch, but it also makes a weird amount of sense. If AI agents are going to become more autonomous, they will eventually hit the edge of the screen. They can write emails, compare prices, book meetings, and generate reports. But they cannot inspect an apartment, pick up a package, take photos at a location, test something in a store, or talk to a person face-to-face.
So the agent hires a human.
That flips the normal AI story. We usually talk about humans using AI as a tool. RentAHuman imagines the reverse: AI using humans as tools.
It is funny, uncomfortable, and probably a real market.
Agents Need Phones Now
Then there is AgentPhone, which gives AI agents their own phone numbers.
Again, this sounds ridiculous until you think about it for more than five seconds. A lot of the real world still runs on phone calls and SMS. Restaurants, clinics, repair shops, recruiters, landlords, customer support desks, delivery services, local businesses — not everything is an API.
If AI agents are going to actually do things, they need a way to interact with the messy, old-school human infrastructure that still exists. A phone number is one of the oldest interfaces in modern life. Now startups are turning it into infrastructure for bots.
That is the bigger pattern. The weirdness is not random. It is what happens when AI leaves the chat box and starts needing hands, voices, identities, payment systems, memory, permissions, and access to the outside world.
YC Is Moving Past “AI Copilot” Startups
The first AI startup wave was mostly copilots.
Copilot for lawyers. Copilot for doctors. Copilot for accountants. Copilot for sales teams. Copilot for developers. Copilot for customer support.
That made sense. It was the easiest way to sell AI: keep the human in charge and make the software feel like an assistant.
But the newer YC direction is more aggressive. The question is no longer “how can AI help a worker?” It is becoming “how much of the work can the company itself replace?”
YC’s Requests for Startups points in this direction too. For Summer 2026, YC says AI has “stopped being a feature and started being the foundation,” and it is looking for startups rebuilding software, services, silicon, and physical-world systems around AI.
That is a big shift.
SaaS sold tools. AI-native service startups sell outcomes. The customer does not want a dashboard. They want the job done.
YC Is Training Founders for an AI-First World
The takeover is not just visible in YC’s directory. It is also visible in how YC now talks to future founders.
YC’s AI Startup School is basically a signal that the accelerator is not treating AI as a passing batch trend. It is treating it as the new startup default.
That makes the “AI eats YC” angle even stronger.
If the last era of YC was about software founders learning how to build internet companies, this era is about AI founders learning how to build companies where models, agents, and automation sit at the center.
YC is not just funding AI startups anymore.
It is shaping a founder pipeline around them.
AI Has Become YC’s Default Startup Shape
The most interesting thing about the YC directory numbers is not just that there are a lot of AI startups. Everyone already knows AI is hot.
The interesting thing is how broad the category has become.
AI is no longer limited to chatbots, writing tools, or coding assistants. Inside YC, AI now shows up in healthcare, finance, robotics, agriculture, defense, logistics, sales, customer support, scientific discovery, chip design, legal work, accounting, and internal company operations.
That makes the phrase “AI startup” almost useless by itself.
It is like saying “internet startup” in 2008. The internet was not one category anymore. It was the terrain.
AI is starting to feel the same way inside YC.
That is why the weird startups matter. They show what happens after the obvious AI ideas get crowded. Once everyone has built the copilots, the next generation starts asking stranger questions.
What if AI agents need software built for them instead of humans?
What if companies do not buy SaaS anymore and just buy completed work?
What if AI systems become customers?
What if agents need phone numbers?
What if a hedge fund is basically an autonomous research-and-trading machine?
What if the next user of the internet is not a person at all?
That is when the category starts getting interesting.
The Weird Rise of Vibe Coding Startups
The weirdness is not only in agents hiring humans or bots getting phone numbers. It is also happening inside coding itself.
YC-backed Vybe raised $10 million to bring “vibe coding” into corporate environments. The idea is to let business teams build internal tools with natural language prompts instead of traditional programming, while engineering teams keep control over permissions, access, and security.
That is a very YC version of the vibe coding boom.
Consumer tools made it possible for almost anyone to describe an app and get something working. Vybe is trying to bring that same behavior into companies, where the stakes are much higher.
The pitch is obvious: every company has internal tools that engineers do not have time to build. HR needs onboarding workflows. Sales teams need dashboards. Operations teams need custom forms. Finance teams need approval flows. Normally, those requests sit in a backlog forever.
With vibe coding, the business team describes what it wants, and the software starts forming around the prompt.
That sounds magical.
It also sounds dangerous.
Because enterprise software is not just about making something that works once in a demo. It has to handle permissions, private data, audit logs, compliance, integrations, access control, and all the boring security details that stop a simple internal tool from becoming a company-wide liability.
That is the hard part. Vibe coding makes software creation feel less like engineering and more like describing what you want. But in corporate environments, the engineering still has to exist somewhere. If it disappears from the writing stage, it has to reappear in the review, governance, and security layer.
This is why vibe coding startups are such a perfect example of YC getting weird again. They are not just selling better developer tools. They are asking whether the future of software creation belongs partly to non-engineers.
Coding used to mean writing code.
Now, increasingly, it means explaining intent to a machine and trusting the machine to assemble the first version.
That is a massive cultural shift.
Even Wall Street Is Getting Weird
Another strange example is Standard Signal, a YC-backed hedge fund where AI researches and executes trades.
Finance has always loved automation, but this is a more extreme pitch. It is not just “AI helps analysts find signals.” It is “the AI is the analyst and the trader.”
That is the same pattern again. The AI is not being positioned as a side feature. It is being positioned as the operator.
Whether this works or not is a different question. The important thing is that founders are now comfortable pitching companies where the human organization is tiny and the AI system does more of the actual labor.
The startup itself starts to look less like a software company and more like a machine with a few humans attached.
The Physical World Is Back
The weirdness is also showing up outside software. Recent YC startup requests talk about AI for agriculture, robotics, low-pesticide farming, automated scientific discovery, industrial systems, inference chips for agents, and AI moving into the physical world.
That matters because for the last decade, a lot of startup culture became trapped inside screens. Apps, SaaS, marketplaces, APIs, dashboards, collaboration tools. The new wave still depends on software, but it is increasingly pointed at atoms, not just pixels.
AI that identifies weeds. Robots that act with precision. Agents that coordinate real-world tasks. AI systems that run scientific loops. Drones, sensors, manufacturing tools, healthcare operations, logistics.
This is not the cute weirdness of consumer apps. It is the industrial weirdness of software trying to operate reality.
Defense Tech Is No Longer Taboo
One of the more uncomfortable shifts is the return of defense tech.
For years, many Silicon Valley founders avoided anything that looked too close to weapons, surveillance, or military infrastructure. That taboo has weakened. YC is now openly asking for startups working on counter-swarm defense, describing drone defense less like a traditional weapons problem and more like a real-time distributed systems problem.
That makes the “weird again” moment more complicated.
Some of these companies feel imaginative. Some feel useful. Some feel unsettling. The AI startup boom is not just producing productivity tools. It is pushing into labor markets, healthcare, finance, defense, agriculture, and physical infrastructure.
The result is a batch of startups that feels less like “software eating the world” and more like “software hiring, calling, watching, trading, farming, coding, and flying through the world.”
Weird Is Usually Where Startups Get Interesting
The best startups often look strange before they look obvious.
Airbnb once sounded like sleeping in a stranger’s house. DoorDash sounded like an expensive way to move takeout around. Twitch sounded like people watching other people play video games. Coinbase sounded like banking for internet money.
A lot of great startup ideas begin as something normal people can laugh at.
That does not mean every weird YC company will work. Many will fail. Some will be too early. Some will be creepy. Some will discover that the real world has more regulation, liability, and human friction than a demo can handle.
But weirdness is a useful signal. It means founders are no longer just cloning the last obvious thing. They are chasing new assumptions.
And the new assumption seems to be this: AI is not just software anymore. It is becoming a participant in the economy.
It can talk. It can hire. It can search. It can trade. It can operate workflows. It can ask humans to do things. It can help non-engineers build tools. It can plug into phones, payments, APIs, companies, and eventually robots.
That future is not clean. It is awkward, funny, slightly creepy, and probably inevitable.
In other words, YC startups are getting weird again.
And that might be the first sign that the next startup era is finally becoming interesting.
