YC’s Fall 2026 RFS: All 13 Ideas, the Army Secretary in the Pitch Room, and the 5 Plays Indie Hackers Can Ship

The Summer 2026 RFS was hard-tech maximalism — 15 categories, 8 of them requiring serious capital: lunar manufacturing, inference silicon, counter-drone defense. Fall trims the list to 13 and quietly swings the pendulum back. By our count, roughly 10 of the 13 ideas are software-first. The banner thesis is still physical — “AI is moving into the physical world” is the opening line — but the physical asks got fewer and weirder (offshore compute flotillas, sensor data collected by robots and balloons), while the software asks got more specific and more shippable.

Two things in this list have never happened before. First, a sitting US government official authored an RFS entry: Army Secretary Daniel P. Driscoll wants founders building “low-cost interceptors,” next-gen sensors, drones, and resilient logistics for ground combat. Second, YC institutionally endorsed small software — not as a stepping stone to scale, but as the point. Koomen's category asks for infrastructure to deploy “purpose-built tools that will only ever have one or a small handful of users.” That is the micro-SaaS thesis, written into the wishlist of the world's most famous accelerator.

One pattern note: Max Kolysh has his name on two entries — AI for the Aging Population and Proving You're Human. Both are demographic bets: one on the age wave, one on the deepfake wave. When a partner doubles up, read it as conviction, the way Diana Hu's triple appearance signaled the agent-infra bet in Summer.

Put the 13 ideas side by side and a strange shape appears. On one end: interceptors for ground combat, data-center flotillas in international waters, robots and balloons collecting “dense physical-world data,” operating systems coordinating robots and human workers on job sites. Nation-scale, capital-heavy, physical. On the other end: a tutor for one child, a care companion for one aging parent, a cloud for software with one user, an agent that quietly fixes one broken API integration. Person-scale, capital-light, intimate.

What is missing is the middle — the horizontal, per-seat, team-collaboration SaaS that defined the last 15 years of startup formation. Not one of the 13 entries asks for it. That is the second consecutive RFS where classic team SaaS is essentially absent, and it confirms the thesis we flagged in April: software is now the substrate, not the moat. The defensible ends of the barbell are physics and regulation on one side, and personal context and distribution on the other. The undefended middle is where AI-assisted competitors clone you in a weekend.

The Driscoll entry deserves its own paragraph, because an RFS co-authored by the Secretary of the Army is not a content-marketing stunt — it is procurement policy showing up in a startup accelerator. “Warfare is at an inflection point, and the old ways of Army acquisition simply don't keep pace with modern threats,” Driscoll writes, calling for “hungry, innovative founders” to build commercially developed, modular systems. The Army has also launched xTechDisrupt, a pitch competition run with YC. For most Superframeworks readers this is not a category to build in — ITAR, clearances, and hardware gate it hard — but it is a loud signal that the US government now treats seed-stage startups as a real defense supply chain.

Four ideas grade High on our tiny-team fit, plus one sleeper where the software slice is wide open. Each comes with a concrete read on what to build first.

The remaining eight ideas split into consumer moonshots — software-first but gated by distribution — and the hard-tech pile, where the gate is capital. Read them as a market map, not a backlog.