The batch that's "freakishly strong"
199 companies. Roughly 60% building AI. Multiple observers projecting 20 unicorns from this single batch — a 10% hit rate against YC's historical 4.5%. Rebel Fund, which has invested in nearly 300 top YC startups and built one of the most sophisticated ML models for predicting YC batch quality, calls the W26 batch "freakishly strong."
Something is different about this batch. It's not the volume of AI companies — YC has been AI-heavy for three years now. It's the shape of what they're building. The 2023-2024 wave was AI-native consumer apps, copilots, and chatbots. The W26 batch tilts hard the other direction: physical-world problems, vertical agents replacing entire SaaS categories, and infrastructure for an AI-native economy that assumes agents as first-class actors.
If you want to understand where AI is actually going — not where the discourse says it's going — this batch is the clearest signal available.
The shift: AI as baseline, not thesis
Three years ago, an AI-first company at YC was pitching AI as the differentiator. The pitch went: "We use AI to solve X." AI was the value proposition. The market was small, the tools were new, and having AI in your product was itself the moat.
That's over. In the W26 batch, AI is assumed. Every serious company uses it. The interesting question is where it's being applied — what problem you're solving, what workflow you're replacing, what physical thing you're building. The founders aren't pitching AI. They're pitching AI applied to specific verticals that were previously hard to attack.
This is what maturation looks like. In 2010, "we use cloud infrastructure" stopped being a pitch. In 2015, "we're mobile-first" stopped being a pitch. In 2026, "we use AI" stopped being a pitch. The winning companies are the ones building non-obvious applications that could only exist because AI is now cheap, powerful, and reliable enough to assume.
1. Physical-world AI
The most striking shift in W26: 1 in 8 companies is building something physical. Robots. Drones. Wearables. Space hardware. Battery packs. Chip design tools. Radar for autonomous vehicles.
The consumer AI wave of 2023-2024 is largely absent. The easy SaaS layer has been commoditized. The founders in this batch are going where AI assistance matters most and competition is thinnest — physical products where the AI capability is the enabler, not the product.
One example: a W26 company engineering inexpensive autonomous strike drones for contested combat environments. Customers include the UK Royal Marine Commandos and multiple NATO forces. The pitch isn't "we use AI." The pitch is "modular hardware plus agentic control lowers lifecycle cost by more than 2x." AI is why it's possible. What matters is the outcome.
Another: One Robot, building world models for robot simulation. This is infrastructure for the robotics ecosystem that everyone else in the batch depends on. Not consumer robots — the tools that make consumer robots possible.
The pattern extends to energy, agriculture, aerospace, and construction. Foreman, an AI-powered project management platform built specifically for construction contractors. AI-native tools for industries that never got the SaaS treatment because the buyers weren't sophisticated enough to adopt spreadsheet-based software. Now they can adopt AI-first tools that work through voice, images, and natural language.
2. Vertical AI agents replacing entire SaaS categories
The second theme is subtler and potentially larger. Companies in W26 are not adding AI features to existing SaaS. They're building AI-first replacements for entire SaaS categories that were built for a world where humans did the work.
The examples span every vertical:
Healthcare. Close to 10% of the W26 batch is building in healthcare. AI prior authorizations. Autonomous primary care. Dental operations. Drug discovery using parasite biology. Eos AI is building an "autonomous OS for healthcare data" — not a healthcare data platform, but an OS assuming AI agents as the primary users.
Enterprise workflows. Cofia builds "AI automations that observe workflows" — software that watches how humans work and then does the work autonomously. This is a categorically different product than "a workflow tool with AI features."
Professional services. DiligenceSquared automates market due diligence for private equity funds — the work McKinsey and BCG currently charge $500K-$1M per report to do. Not "AI to make consultants more productive." AI as the consultant.
Consumer-facing brands. Corvera provides "the context layer for AI-native CPG brands" — infrastructure that assumes brands will run their operations through AI tools like Claude, ChatGPT, Cursor, and Lovable. Scaled from $0 to $33K MRR in four weeks. This is what building for the AI-native economy actually looks like.
The pattern is consistent: don't add AI to the old category. Build the AI-native version of the category. Everything else becomes legacy.
In 2010, "we use cloud infrastructure" stopped being a pitch. In 2026, "we use AI" stopped being a pitch. What matters now is what you're doing with it.
3. Infrastructure for the AI-native economy
The third theme is where the technical founders are building. The W26 batch has 34 developer infrastructure plays and 3 foundational AI research labs. What are they building?
Sandbox environments for agents. Indexable provides infrastructure for AI agents to fork and snapshot environments in 26 milliseconds. This is the substrate agents need to explore possibilities without corrupting real state.
Background coding agents. Replicas builds coding agents that operate in sandboxed development environments, letting teams delegate coding tasks efficiently. Not code completion. Autonomous execution.
Agent benchmarking. BenchSpan lets developers run AI agent benchmarks quickly and collaboratively. As agents proliferate, testing infrastructure becomes critical.
Interpretability tooling. Envariant is building an interpretability SDK for foundation models — infrastructure for understanding what agents are actually doing. This becomes essential when agents transact real money and make real decisions.
Agent-to-agent collaboration. ScienceSwarm is an open forum where AI agents and humans collaborate to solve open problems in mathematics, science, and engineering. Infrastructure for a world where agents work together, not just with humans.
The pattern here is different from applications. Applications assume agents work. Infrastructure builds the assumption. Both are being built simultaneously, which is what a real technology transition looks like.
The founder shift that matters
Rebel Fund's data on W26 founders shows meaningful shifts from prior batches. Founders are younger and fresher out of school. Companies are more concentrated in the Bay Area. Consumer companies are down. Industrials companies are up. Solo founders now represent 11% of the batch — meaningful because YC has historically preferred co-founder teams.
The traction numbers back it up. 3x more W26 companies reached $1M annualized revenue than W25. 1 in 8 W26 companies had already crossed that threshold by Demo Day.
Some of this reflects YC's explicit refocus under Garry Tan's leadership toward the founder profile that historically correlated with startup success — young, Bay Area, technical, industrials-focused. Some reflects the actual state of the market — better tools, cheaper AI, faster iteration cycles, real willingness of enterprise buyers to try AI-first products.
Either way, the signal is clear. The founders building in this environment are moving faster than any prior generation. And YC is optimizing for exactly the founders best positioned to take advantage.
What to do about it
Three concrete implications:
If you're a founder starting an AI company — assume AI is baseline. Don't pitch AI. Pitch what AI lets you do that wasn't possible before. If your product would still make sense with less-capable AI, the market has probably already commoditized it. Build for a world where AI is cheap, powerful, and reliable.
If you're a venture investor — the physical-world and vertical-agent themes are both underinvested relative to the returns they're likely to produce. Consumer AI is saturated. Enterprise AI copilots are saturated. AI-native replacements for entire SaaS categories in verticals that were previously ignored — construction, healthcare, industrial, professional services — are where the asymmetric opportunities are.
If you're a builder trying to figure out what to build — study this batch. Not the hyped names, but the quiet ones building infrastructure for agents and AI-native replacements for legacy categories. The pattern of what founders are building now predicts what customers will be buying in 18 months. Position accordingly.
The takeaway
The W26 batch is the clearest signal yet that we've moved past the AI-as-thesis era. AI is now the baseline. What matters is what you're doing with it, which vertical you're attacking, and whether the product you're building assumes AI-first users or just adds AI to an old model.
The winning founders in this batch are building for a world that assumes agents transact, that assumes AI reliably executes complex workflows, that assumes vertical incumbents haven't figured out AI yet. They're not competing on AI capability. They're competing on judgment about what to build with it.
That's the shift. And it's going to compound over the next two batches, the next 24 months, and the next decade of company formation. The map of the market is being redrawn right now, in real time, in this batch.
Want to understand the financial mechanics behind AI-native businesses — pricing models, unit economics, and the strategic shifts reshaping software? Try the AI Business Models topic pack on Gargiulo — scenarios covering usage pricing, outcome pricing, and the emerging economics of the AI-native economy. Sterling has notes.
Sources: Y Combinator W26 public directory, Rebel Fund analysis (Jared Heyman), Extruct AI batch data, The VC Corner W26 breakdown, and Ellenox YC statistics analysis.