On a mid-March afternoon in San Francisco, roughly 1,500 investors filed into Y Combinator's Demo Day—the biannual ritual where Silicon Valley's most influential accelerator unveils its latest class of startups. This time, though, the usual cross-section of software ideas and oddball hardware felt notably lopsided.
AI video generators. Agent orchestration platforms. Tools to automate Zoom calls, restaurant scheduling, even university grading. By one investor's rough count posted to LinkedIn afterward, some 133 of the batch's 160-odd companies touched machine learning or artificial intelligence in some capacity.
Not a trend. A thesis.
The Winter 2025 cohort looked less like a diverse portfolio and more like a concentrated bet on a single technological shift. Perhaps the most telling entry: Swerve, a consumer app pitched as "Instagram, but for AI-generated videos." Whether that's visionary or a sign the space has gotten too crowded depends on whom you ask.
Four Batches, One Direction
YC announced the shift to quarterly cohorts in January, doubling its previous cadence. The W25 program ran from January through March, following application deadlines set the prior November. Standard terms applied: $500,000 total per company, split into a $125,000 investment for 7% equity via post-money SAFE, plus $375,000 on an uncapped most-favored-nation SAFE.
The logic behind the expansion was straightforward enough—more shots on goal, broader surface area across emerging categories. But the composition of W25 suggested something else was happening. YC wasn't just adding capacity. It was loading up on a particular kind of founder: those building for a world where generative AI has already won.
Third-party catalogs like GPAgent tallied 167 companies in the batch, though YC itself hasn't released official numbers. Demo Day unfolded across two days, March 12–13, with the usual tight schedule and back-to-back pitches that leave little room for nuance.
Swerve's Social Gambit
Jason Zhou and Soroush Saryazdi, Swerve's co-founders, are betting that AI video generation doesn't need to be a professional tool. Their pitch: turn it into something casual, social, shareable. "Snap AI videos with friends," as the tagline goes.
The iOS app launched with a demo that emphasized speed—seconds, not minutes, to generate clips. Details on user traction or engagement metrics were scarce at Demo Day, which isn't unusual for early-stage pitches. What stood out was the framing. Most startups in the AI video space are chasing creators, marketers, enterprise buyers. Swerve is aiming for something closer to early Instagram: lightweight, native mobile, built for distribution rather than production quality.
It's an interesting gamble. Social apps live or die on virality and sustained engagement, neither of which generative video has proven capable of driving at scale—yet. The tech is still uneven. Output can veer from impressive to uncanny within the same session. But Swerve's presence in the batch signals that YC, at least, believes the infrastructure is close enough to bet on consumer adoption.
Or maybe the accelerator is just flooding the zone.
The Agent Economy, Assembled
AI video wasn't the whole story. TechCrunch's March 13 roundup of the batch highlighted a broader push into agent tooling and infrastructure—the pipes, not just the apps.
Pickle claimed 1,500 paying users for its "AI clone for Zoom calls," a pitch that sounds absurd until you consider how many hours of video meetings professionals now endure weekly. Rebolt built AI agents for restaurant management and reported being in pricing talks with Restaurant Brands International, Burger King's corporate parent—a signal that enterprise buyers are at least entertaining these tools, if not yet committing budgets at scale.
Then there was Abundant, which developed an API for agent teleoperation, essentially handing off tasks between AI and humans mid-workflow. Browser Use went open source with tools designed for web-browsing agents. The category felt less like a single product thesis and more like infrastructure plays for a market that might materialize in 18 months—or might not.
The batch also included wild cards that had nothing to do with software. Red Barn Robotics pitched an autonomous weeding robot and claimed $5 million in letters of intent for the upcoming growing season. Splash, working on autonomous patrol boats, said it had logged over 200 miles in San Francisco Bay. GradeWiz aimed to automate university grading. Retrofit used AI to curate vintage clothing marketplaces, which seems like either a brilliant niche or a solution in search of a problem.
The spread suggested YC isn't particularly interested in picking category winners. It's placing a lot of small bets and hoping a few break through.
Six Months Later, a Seed Round
Artificial Societies, another W25 company, raised a $5.35 million seed round in August—six months post-Demo Day—led by Point72 Ventures. The startup simulates audiences and societies using agent-based models, a concept that sits somewhere between academic research and commercial product.
CEO James He described it as a way to test marketing campaigns or policy decisions before committing real-world resources. "You can run a thousand variations in a day," he told reporters at the time. It's the kind of abstraction that makes venture capitalists lean in and everyone else squint a little.
But that abstraction—agents simulating people, models generating content, AI handling workflows previously reserved for human judgment—ran through much of the batch. YC's expanded schedule gives the accelerator more capacity to place bets across categories. W25 looked like the cohort where those bets tilted decisively toward infrastructure and tools that assume AI is already embedded in daily workflows, not some distant inevitability.
What It Means for the Next Wave
For founders watching from the outside, the signal is clear enough. YC is all-in on generative AI, agentic systems, and the infrastructure layer needed to make both work at scale. Whether Swerve becomes the breakout consumer hit or joins the graveyard of short-lived social experiments, its presence in the batch tells you where the money is flowing.
The risk, of course, is that everyone is building for the same future at the same time. When 133 companies in a single cohort are touching AI, someone's going to be wrong about market timing. Or product-market fit. Or whether consumers actually want AI-generated videos cluttering their feeds.
But that's the bet YC is making—saturate the space, move fast, hope a few break out before the capital window closes. Winter 2025 might eventually be remembered as the batch where the accelerator went all-in on one technology shift. Or as the moment the market got too crowded to sustain that many bets at once.
For now, the factory is running. Demo Day is over. And the real test—whether any of these companies can turn early traction into durable businesses—begins.
