What YC's New 'Requests for Startups' List Says About the Bet on AI-Native

The Request for Startups is the document in which Y Combinator lists, for each batch, the ideas it would like to see founders tackling. YC itself notes that these themes represent only a fraction of what it funds and that no one needs to work on them to apply. Even so, the RFS functions as an allocation signal: it's where the network's partners and founders put in writing where they think money and attention are headed in the coming years.

The Fall 2026 list has a central thesis stated right at the top: "AI is moving into the physical world." The bet is that the next wave of startups will rebuild the systems that run the real world (education, health, defense, finance, infrastructure, and work itself), not just screen-based software anymore. The RFS is worth reading not for the hype, but for what it reveals about market structure. Below, a reading of the categories organized around the question that matters for those building tech businesses in Brazil: can this be replicated here?

The Theses That Assume US Infrastructure or Context

Some categories are non-transferable by design. The Future of American Defense, signed by US Secretary of the Army Daniel P. Driscoll (the first time an RFS has featured a sitting secretary), is literally a Pentagon shopping list: low-cost interceptors, sensors, drones, resilient logistics. The capital funnel and the proving ground are governmental and American. There's no Brazilian equivalent, neither in budget nor in purchasing willingness, which makes this a thesis a founder in Brazil can hardly tackle from here.

Compute at Sea, by Francois Chaubard, proposes moving data centers to the ocean as a response to the lack of power, land, and permitting on dry land. It's an extremely capital-intensive thesis, anchored in regulatory and power-grid constraints specific to the American market. Data for the Real World (Austin Tindle and Diana Hu) and New Operating Systems for the Physical World (Charlie Warren) point to dense collection of physical data and systems that orchestrate agents, robots, and humans in the field. The opportunity exists in Brazil (agribusiness and logistics are cited as among the largest industries in the world), but the barrier here is field robotics and cheap sensors at scale, a hardware game that requires patient capital that Brazilian venture capital rarely provides.

The Theses That Already Exist or Make Sense in Brazil

Two categories stand out because YC itself explicitly cites Latin America. In The Best Time to Build in Crypto, Nemil Dalal argues that the bear market is the best time to build and names BlindPay and Infinia specifically as projects "building the developer interface and ramps for Latin America," alongside Aspora, which focuses on transfers to India. The angle here is concrete: stablecoin rails for remittances and payments in markets with volatile currencies and expensive banking systems. This is territory where the Brazilian founder has a contextual advantage, not a disadvantage. The thesis of "agentic commerce" and stablecoins as the financial rail for AI agents is global, but the currency exchange and remittance problem is more acute (and therefore more valuable to solve) in the Global South than in California.

AI-Native Compliance Infrastructure, by Daivik Goel, describes financial compliance still "stitched together with spreadsheets." The text talks about state-by-state licensing in the US, but the logic carries over: Brazil has its own regulatory patchwork (Banco Central, Brazil's central bank, the CVM, Brazil's securities regulator, accessory tax obligations, and the tax complexity that spawns entire companies just to deal with it). An AI-native compliance infrastructure for Brazil's regulatory reality is a thesis that makes local sense, even if the addressable market is smaller.

The Horizontal Theses, Where Brazil Competes on Equal Footing (or Doesn't)

Some categories are pure software and, in theory, have no borders. A Cloud for Small Software, by Pete Koomen, bets on "small software": tools with one or a few users, generated by agents, that today are easy to build but hard to deploy and share. The analogy is being "as easy to share as a Google Doc." Multiplayer AI, by Aaron Epstein, starts from the observation that Figma and Google Docs won by going multiplayer, while AI use is still single-player: you open a chat, type, and get a reply in a box only you see. The thesis is collaborative agents by default, where the whole team joins the same live session.

The honest counterpoint: horizontal developer-tool theses are exactly where the Brazilian founder competes at a disadvantage. The buyer is global, but proximity to the early-adopter customer, access to American Series A capital, and the density of talent around frontier models are all in San Francisco. Winning in "small software cloud" from Brazil is possible, but in practice it requires distribution and go-to-market built for the US market from day one, not as a future expansion.

Self-Maintaining APIs, by Harsha Gaddipati, is the most concrete item on the list. The idea: when Stripe publishes a breaking change, an agent should scan customers' code, identify the affected usages, and open a PR with the fix, something like "Dependabot for APIs." The author cites that at AWS more than 30% of service downtime came from unnoticed external API/package changes. This is a dev infrastructure thesis that's geography-independent, where the barrier is technical, not market-based.

The Consumer Theses and the Mission-Driven Ones

AI-Powered Consumer Products for 1 Billion People, by Raphael Schaad, brings the most quotable number in the RFS: the "magic" that today costs about $1,000/month in tokens per user would be falling 10x a year. If the curve holds, the mass-consumption moment arrives soon. AI for the Aging Population (Max Kolysh) notes that by 2030 one in five Americans will be over 65, with millions of caregiving positions unfilled and 53 million family members already caregiving for free. The Primer (Andrew Miklas) describes an AI tutor to teach children to read, write, and do math with the quality of a private tutor.

These consumer theses have a trap for Brazil: the token economics Schaad describes depends on inference costs falling, but Brazilian consumer purchasing power and the ticket size that sustains CAC are much smaller. A consumer AI thesis that pencils out in the US might not pencil out here. Proving You're Human (Max Kolysh), on the other hand, about rebuilding the internet's trust layer against deepfakes (the piece opens with the case of the finance employee who wired $25 million on a call where everyone else was a deepfake), is a global and geographically neutral problem.

What the Brazilian Founder Takes Away From This

The RFS is a map of where American capital wants to go, not a mandate. The useful reading isn't "I'll build what YC asked for," but understanding that the stated axis is AI moving from the screen into the physical world and into critical infrastructure. Within that, the most defensible space for those building from Brazil is where local context becomes an advantage: stablecoin and remittance rails (which YC itself already funds in LatAm), AI-native compliance for Brazilian regulation, and borderless dev tools like self-maintaining APIs. The defense, ocean-compute, and billion-scale consumer theses are, in practice, hard to attack from here, whether due to capital barriers or market barriers. A YC round is a means, not an achievement, and the RFS is worth more as a reading of thesis than as a shopping list.