Table of Contents

• First, how to read an RFS without fooling yourself
• The requests that translate well to Latin America

  • Crypto as payment infrastructure, not speculation
  • Compute, data, and agents for industries that don't live at a desk
  • Physical-world data
  • Proving you're human
  • AI-native compliance infrastructure
• The ones that sound amazing and probably aren't for you
• The request that matters most for the region
• What to do on Monday
• FAQ

First, how to read an RFS without fooling yourself

There's a trap in these lists, and it's worth naming before getting excited.

An RFS is not a validated idea. It's an investor's hypothesis, written in two paragraphs, without having talked to a single customer. YC says so explicitly: these represent a fraction of what they fund, and you don't need to be working on any of them to apply. The Fall 2026 list includes a request for compute floating in the ocean. The author himself writes "this sounds crazy." That's honesty, not validation.

The second trap is more subtle. These requests are written from a U.S. context, with assumptions that don't travel: abundant capital, willingness to pay in dollars, homogeneous regulatory markets, enterprise customers who buy software through Stripe without calling legal. Copying the idea without translating the context produces startups that solve a problem that doesn't exist in your market, or that exists with a completely different economics.

The right way to use the list is as a diagnosis, not a menu. Each request describes a structural shift — AI got good enough, sensors got cheap enough, regulation finally became clear. That structural shift is real and global. The specific solution YC proposes is just one bet. Your job is to take the shift and ask how it shows up where you live.

The requests that translate well to Latin America

Crypto as payment infrastructure, not speculation

This is the most direct one, and YC almost wrote it with the region in mind. Nemil Dalal argues that this is the best time to build in crypto precisely because prices are down and the opportunists have left. Bear markets let you build without competing against someone promising impossible returns.

What matters for us is in the examples he cites: BlindPay and Infinia building developer interfaces and on/off ramps for Latin America. Aspora doing remittances to India. In other words, YC's thesis on crypto in 2026 isn't DeFi or NFTs — it's stablecoins working as payment rails for markets where the traditional financial system is expensive, slow, or inaccessible.

That's literally our problem. An Argentine freelancer billing a client in Berlin, a Colombian SME paying suppliers in Shenzhen, a remittance from Miami to Guadalajara: each of those flows loses between 3% and 8% to friction. The advantage of building this from the region is not small — you understand why people distrust the bank, you know the local regulator, you know adoption is won through WhatsApp, not API documentation.

If there's one request on this list where a Latin American founder has a structural edge over one in San Francisco, this is it.

Compute, data, and agents for industries that don't live at a desk

Charlie Warren opens his request with a stat that reframes things well: 80% of the global workforce doesn't work sitting in front of a computer. And the software for those people — construction, maintenance, fleets, logistics — hasn't changed much in twenty years. It dispatches people, tracks them, manages assets, bills the client.

What's changing now is that three kinds of workers coexist: AI agents that quote complex jobs and assemble crews, robots deployed in the field, and humans wearing devices that log everything they do. No current operating system was designed to coordinate all three. How do you route a job between an agent, a robot, and a person? What does "safety" mean when humans and robots work side by side?

The economic argument is strong: these industries spend 10 to 100 times more on labor than on software. Whoever coordinates the labor — not just visibility into it — captures a much bigger market than the current one.

In Latin America, the robotics part is far off, let's be honest. But the part about agents coordinating human field work is fully viable today, and the industries exist at scale: construction in Mexico, agriculture in Brazil and Argentina, mining in Chile and Peru, logistics everywhere. The regional version of this request probably won't include robots for several years. It does include an agent that quotes, schedules, verifies, and invoices work that today gets coordinated through WhatsApp groups and spreadsheets.

And there's a second prize, which Warren points out well: by running the system, you end up recording how work actually happens in reality. Nobody else has that data — not the model labs, not the robotics startups, not the incumbent software.

Physical-world data

Austin Tindle and Diana Hu describe the same gap from another angle. Models are superhuman at code, language, and images, because that data was abundant on the internet. Physical-world data, by contrast, is scarce, captured by sensors designed to be read by a human. With better foundation models and increasingly cheap sensors, collecting dense physical-world data is finally viable.

The examples they give: Gecko Robotics sending robots into inaccessible places to build predictive models, and Sorcerer using autonomous weather balloons to feed U.S. government forecasts. The logical chain is simple and powerful: more real data enables precise modeling, and what can be modeled can be controlled.

The world's biggest industries — energy, agriculture, logistics, construction — currently operate on limited data and intuition-based models. In Latin America that's even more true, and that's where the opportunity lies: we're one of the most important agricultural and mining regions on the planet, with low instrumentation density. You don't need to invent the sensor. You need to deploy it, capture the data nobody's capturing, and build the model on top.

Proving you're human

Max Kolysh opens with the case of the finance employee who joined a video call with his CFO and several colleagues, and wired $25 million. Everyone else on the call was a deepfake.

The underlying problem: every trust signal we use was designed for a world where faking a human was expensive. That world is over. Seeing a face or hearing a voice no longer proves anything, and fraud is exploding.

Rebuilding the internet's trust layer — knowing there's a verified human on the other end of a call, a message, a transaction, ideally without forcing everyone to hand over their privacy — is one of the most important problems of the decade. And it's not just anti-fraud: it's Twitter without bots in the replies, dating apps where every match is real, reviews written by people who actually bought the product.

This is a global problem, but Latin America has a particularity that makes it urgent: high digital banking and fintech penetration, mass adoption of instant payments like Pix, and a sophisticated fraud ecosystem already operating at scale. The need arrived before the solution.

AI-native compliance infrastructure

Daivik Goel describes financial compliance as something stitched together with spreadsheets, siloed tools, and expensive headcount. Every new market multiplies the complexity, and the cost of staying compliant grows faster than revenue.

The argument for why this is a natively AI problem is convincing: most compliance work is monitoring regulatory changes, flagging anomalies, generating reports, and maintaining audit trails. Tasks a model does faster and cheaper than a person. And yet almost every current solution is still built around manual workflows with human review as the bottleneck.

If the "state-by-state licensing in the U.S." example sounds like it doesn't apply here, look again. A fintech operating in Mexico, Colombia, and Brazil faces three distinct regulatory regimes, in two languages, with renewal and audit cycles that don't talk to each other. The Latin American version of this request is probably more painful than the original.

The ones that sound amazing and probably aren't for you

Not every request is equally accessible, and confusing them is expensive.

The future of American defense, signed by the U.S. Secretary of the Army, is an explicit request for low-cost interceptors, sensors, drones, and advanced manufacturing for ground combat. It's the first time a sitting official has written an RFS. It's fascinating as a signal of where capital is moving, and it's practically inaccessible to a Latin American founder for reasons of citizenship, security clearances, and government contracting.

Compute at sea: modular fleets of ships operating as a global cloud, taking advantage of the fact that the ocean covers 70% of the earth's surface, requires no permits, and works as a natural heat sink. The logic is real: data centers are running out of electricity and land, and communities are pushing back. But it's a capital-intensive hardware bet, and that's not the best first company for someone without access to deep venture capital.

Consumer products for a billion people deserves a finer distinction. Raphael Schaad's argument is that intelligence is already good enough to treat an agent like a person, and the cost is falling 10x per year — today the magic might cost a thousand dollars a month in tokens per user, tomorrow it won't. Whoever builds now keeps the moment.

That cost curve is exactly the argument for why this could be born in Latin America instead of San Francisco. A product that needs to charge $30 a month to be viable in the U.S. has to work at $3 here. Building under that constraint is harder today and a huge advantage once costs fall. It's not impossible; it's demanding.

The request that matters most for the region

Of the thirteen, the one that stands out most for Latin America is The Primer, from Andrew Miklas.

The reference is to Neal Stephenson's The Diamond Age: an interactive book that adapts completely to a girl and, through stories tuned to her life, teaches her not just to read but to think and reason. It grows with her over years.

The underlying observation is that the best education has always come from one-on-one tutoring. Aristotle taught Alexander. That privilege was reserved for very few. What YC is asking to be built today isn't the full Primer, but something concrete: a product that adaptively teaches young children to read, write, and do arithmetic with the quality of a dedicated private tutor, at consumer scale. Not to replace teachers, but to make them more effective.

Now translate that to a region where standardized tests systematically show that a majority of ten-year-olds can't comprehend a simple text. Where a private tutor is a luxury of the urban upper class. Where the phone is already in mom's pocket.

The market isn't "parents who want a competitive edge for their kid." It's "parents who see their kid isn't learning to read and have nowhere to turn." It's a bigger, more urgent problem, with less competition, than the U.S. version of the same idea.

What to do on Monday

If any of this moved you, the worst possible reaction is to open a document and start designing the product.

YC's list is a hypothesis about what changed in the world. Your job isn't to adopt their solution — it's to verify whether the change they describe has already reached your market, and what shape it took getting there. That gets verified by talking to people, not by reasoning about it.

Take the request that moved you most and answer three questions before writing a single line of code:

• Who has this problem today in your country, by first and last name, not as a demographic category.
• What is that person doing right now to solve it, because they're always doing something, even if it's the wrong thing.
• What changed in the last eighteen months that makes a solution possible that wasn't before. If you can't answer this one, you're probably chasing an old idea with new vocabulary.

After that, if the answer still holds, build. And apply. YC says it at the end of almost every request, in the same words: we'd love to hear from you.

FAQ

What is a Y Combinator Request for Startups (RFS)? It's a periodic list where YC partners describe the kinds of startups they'd like to see get built. It's not a requirement to apply to the program, nor a customer-validated idea — it's the fund's read on where market space is opening up.

Why is the central theme of the Fall 2026 edition AI in the physical world? Because, according to YC, the phase of applying AI to pure software (chat bolted onto a SaaS) has matured. The next leap is rebuilding systems that operate in the real world: field work, sensors, health, defense, finance, and education.

Which of the thirteen requests makes the most sense for a founder in Latin America? Crypto as payment infrastructure is the best fit, because the region already lives the currency and financial friction problem that request aims to solve. AI-native compliance for multi-country fintechs and agents that coordinate field work are also strong candidates.

Which requests on the list are unrealistic to build from the region? Defense (due to citizenship and government contracting restrictions) and compute at sea (due to capital-intensive hardware) are the least accessible for a first project without deep venture capital.

---

Source: Requests for Startups — Y Combinator, Fall 2026 edition.