The list nearly doubled and its centre of gravity moved from the laptop to the factory floor in a single quarter. That is not a gradual drift in taste. That is a thesis being rewritten in real time, and each entry is signed by the partner who will actually evaluate the applications — Garry Tan on agriculture, Tyler Bosmeny on counter-swarm, Diana Hu on semiconductor supply chains, Adi Oltean on the moon.
An RFS isn't a prediction. It's a commitment. The people who wrote it are the people writing the cheques.
The line that should stop you
The mandate underneath all fifteen categories is that software is now the substrate, not the moat.
Sit with that for a second, because of who is saying it. This is the institution that funded Airbnb, Stripe, Dropbox and Coinbase — built entirely on the premise that two people with laptops could out-execute anyone. YC is now saying, in its own document, that writing the code is no longer the hard part.
Which follows logically, if uncomfortably. If AI can generate software, software stops being scarce. What stays scarce is everything AI can't shortcut: regulatory approval, physical supply chains, radiation tolerance, institutional trust, a robot that actually works in a field.
Tyler Bosmeny's framing of counter-drone startups is the sharpest version — he wants companies that look like Cloudflare, not Raytheon. Software-defined defence at a fraction of missile-system cost. The software is assumed. The wedge is the physical domain it's being pointed at.
The cluster nobody is counting
Your list flagged the two space categories as a signal that YC is serious about orbit. Fair. But there's a bigger cluster hiding in plain sight.
Four of the fifteen categories are about rebuilding software for agents: Company Brain, The AI Operating System for Companies, Software for Agents, and Dynamic Software Interfaces. (Worth noting — Company Brain and The AI OS for Companies are two separate entries, not one.)
Software for Agents asks founders to rebuild every major software category for a world where the next trillion users aren't people. APIs, docs, identity, permissions, payments — designed for autonomous programs. Dynamic Software Interfaces is its mirror: rebuild interfaces so agents operate software natively instead of scraping tools built for humans.
Space got two categories. Agent infrastructure got four. If you're reading the RFS for where the volume of funded companies actually lands, that's your answer.
Where the thesis strains
Now the honest tension.
Can YC fund this? A tape-out for space-rated silicon costs multiples of $500K before you have a chip. Counter-drone systems live or die on defence procurement cycles measured in years, not twelve-week sprints. The batch model — fast iteration, early users, demo day traction — is optimised for exactly the kind of company this RFS is de-emphasising.
The rebuttal is that money is no longer the scarce input. $297 billion flowed into startups globally in Q1 2026, roughly 2.5× the prior quarter and the largest quarter ever recorded. a16z raised $15 billion. Thrive closed over $10 billion. Accel raised $5 billion. Defence tech alone took a record $49.1 billion in 2025, and Anduril raised $4 billion at a $60 billion valuation in March.
In that environment, YC's product was never really the $500K. It's the signal — and the signal now routes hard-tech founders toward capital that is visibly hunting for exactly this.
My read
The RFS is less a list of ideas than an admission: the garage model produced enormous returns for twenty years and is now, on YC's own account, insufficient. The founders it wants next have to write the code and then build the thing.
I'd take it seriously. I'd also note that a thesis this fast-moving — 8 categories to 15 in a quarter — is a thesis under revision, not a settled one. Check what the Winter list says.
Question for you: if software is the substrate and not the moat, what's the last durable moat left for a company with no factory, no fab, and no regulatory licence? I have a guess — distribution — but I'm not sure it survives agents either.
