Y Combinator's Requests for Startups, Fall 2026: all 13 ideas explained.

YC describes the Requests for Startups as a tradition: ideas it would like to see founders tackle. It is careful to say the list is not a requirement. The ideas are only a fraction of what it funds, a founder excited by one can take that as extra validation, and nobody needs to work on any of them to apply.

The Fall 2026 requests come from YC partners and from founders working at the frontier of their fields. Each request names its author, so read it as that person's view of an opportunity, not as a promise from YC.

The 13 requests in the order YC lists them, as checked on September 13, 2026. The one-line summaries and the likely buyers are ours; each title links to the full request on YC's page.

The Primer

Named for the interactive book in Neal Stephenson's novel The Diamond Age, this request asks for an adaptive tutor that teaches young children reading, writing and arithmetic with the quality of private tutoring and the reach of a consumer product. It pictures a tutor that learns one child over years and supports classroom teachers rather than replacing them. Read Andrew Miklas's request.

Marketing it: the parent buys and the child uses, so the product has to win twice. Parents want evidence that their child is learning, not just spending time on a screen, which points to visible progress, careful claims and word of mouth between parents. The request's point that the tutor should support teachers is also a marketing asset: a product teachers are glad to recommend has a channel parents already trust.

AI-Powered Consumer Products for 1 Billion People

Raphael Schaad argues that AI is now capable and cheap enough to build mass-market products for how people learn, travel, stay healthy, manage their money, play and socialize, and asks for products meant for billions of people rather than a niche. Read Raphael Schaad's request.

Marketing it: a product for everyone still has to find its first thousand people. In most cases that means one clear use and one channel to start, such as app store search, creators or word of mouth, plus a reason for people to share it. Retention decides whether spending on growth pays off, so track whether people come back from the first week.

AI for the Aging Population

Max Kolysh asks for technology built for adults over 65: voice interfaces people can simply talk to, safety monitoring, robots that help at home and software that lets family caregivers coordinate care and emergencies. The request points out how little technology is designed with older users in mind. Read Max Kolysh's request.

Marketing it: in many families the person who pays is an adult child or a caregiver, and the person who uses it is their parent. The message has to respect both: dignity for the user, peace of mind for the family and clear answers about privacy and what happens in an emergency. In most cases trust comes through people families already rely on, such as doctors, senior living communities and caregiver groups.

A Cloud for Small Software

Pete Koomen starts from a shift: AI agents make it easy to build small, custom tools for one person or a small team, but today's big clouds were built for large software projects, with the complexity, sign-in and permissions to match. He asks for a cloud where agents can deploy these tools and people can share them as easily as a document. Read Pete Koomen's request.

Marketing it: the people who benefit may never read a word about infrastructure, so the message has to be about the finished tool: what you can make and who you can share it with. Discovery can come from where these tools get built, inside the AI coding tools and agents people already use, and from a gallery of real examples. Developers who build for others are a second audience with different questions, and in most cases a page that tries to answer both answers neither well.

Multiplayer AI

Aaron Epstein notes that today's AI tools are built for one person at a time, while real work, in engineering, sales, support, legal, analysis and marketing, is done by teams on tasks that run for hours or days. He asks for agent workspaces a team can share live: watching, redirecting and handing work to each other the way colleagues do. Read Aaron Epstein's request.

Marketing it: there are two audiences, the manager who buys and the team who has to adopt it. The strongest proof is the moment a teammate picks up an agent's work halfway through a task, so build demos and short videos around that moment. Starting with one function, such as support or engineering, makes the examples sharper than a product pitched at every team at once.

Self-Maintaining APIs

Harsha Gaddipati describes a familiar pain: API providers ship breaking changes with little warning, new features go unnoticed and changelogs go unread. The request asks for infrastructure that lets the provider's agents update customers' code directly, by opening pull requests. Read Harsha Gaddipati's request.

Marketing it: the buyer is the API company, but the product lands in its customers' code, so trust is the whole sale. Developers judge by documentation, sample pull requests, security details and what the tool does when something goes wrong. In most cases developer marketing runs on docs, open code and technical writing more than on ads.

AI-Native Compliance Infrastructure

Daivik Goel's request targets the patchwork of state licenses, renewals and international rules that businesses have to keep up with. It asks for compliance systems built on AI that monitor regulations, flag anomalies, prepare reports and keep audit trails, replacing scattered tools and much of the manual work that falls on compliance teams. Read Daivik Goel's request.

Marketing it: compliance buyers are paid to avoid risk, so a new vendor is a risk until proven otherwise. They look for accuracy, audit trails, security reviews and references from companies like theirs. They also search for answers to specific rule questions, which makes accurate, specific content a practical way to be found and trusted before the first sales conversation.

Proving You're Human

The second request from Max Kolysh is about fraud: deepfakes and voice clones make it hard to know who is on the other end of a call, a message or a payment. The request cites a finance worker who wired out $25 million to deepfakes, and asks for ways to confirm a real person is present without making everyone give up their privacy. Read the request.

Marketing it: security buyers hear frightening pitches all day, so the stronger message is evidence: independent test results, how the product handles false alarms and what data it keeps. In most cases there are two groups to convince, the platform that installs it and the honest users who have to pass the check, so the experience for those users belongs in the marketing too.

The Best Time to Build in Crypto

Nemil Dalal argues that a down market attracts builders who care about real use rather than a quick payday, and lists areas to build: rails for raising capital, stablecoins, tokenized securities, payments made by AI agents, trading systems and blockchains built for privacy. The request goes as far as saying YC expects every YC startup to use crypto rails eventually, from raising capital to payments. Read Nemil Dalal's request.

Marketing it: trust is the product. Fintech companies and financial institutions will check licensing, audits and security before they read a feature list, and developers will judge the documentation. Any claim about returns, yields or the safety of funds should be reviewed by a qualified lawyer before it is published, because in finance one careless sentence can become a regulatory problem.

The Future of American Defense

This is the first time the list has carried a request from a sitting US Secretary of the Army. Daniel P. Driscoll asks startups to build modular, commercially developed technology the Army can adopt: low-cost interceptors, next-generation sensors, software, payloads, drones, resilient logistics and advanced manufacturing. He stresses open architectures, equipment that works in conditions from the Arctic to underground, and a pace that traditional acquisition has not matched. Read Secretary Driscoll's request.

Marketing it: the customer is the Army, and buying runs through formal procurement, with evaluators, testing and contracting rules. Marketing's job here is credibility: a clear capability statement, evidence from tests and pilots and a website that answers what an evaluator would check. In most cases demonstrations and relationships carry more weight than advertising.

New Operating Systems for the Physical World

Charlie Warren's request looks at industries such as construction, maintenance and fleet operations, which, the request notes, spend far more on labor than on software. It asks for systems that coordinate three kinds of workers at once, AI agents, robots and people wearing devices, and that learn from the complete record of the work. Read Charlie Warren's request.

Marketing it: operations leaders in these industries buy from people they trust and from what they can see working. Site visits, field demonstrations, trade associations and case studies told in their own units, such as hours saved, jobs finished and safety incidents avoided, in most cases do more than abstract claims about AI.

Data for the Real World

Austin Tindle and Diana Hu argue that energy, agriculture, logistics and construction still run on sparse data and intuition. They ask for robots, autonomous devices and sensors that collect dense data about the physical world, so it can be modeled and, eventually, controlled. Read Austin Tindle and Diana Hu's request.

Marketing it: selling data means proving it is accurate and worth acting on. Buyers want to see coverage, error rates and a decision the data changed. In most cases the first customers come from one industry where the gap hurts most, and their results become the proof for the next industry.

Compute at Sea

Francois Chaubard proposes moving data centers onto the ocean: flotillas of modular vessels that work together as a global cloud. The case he makes is that the ocean offers space, sunlight and natural cooling at a time when data centers are short of electricity and land. Read Francois Chaubard's request.

Marketing it: the early customers are few and large, AI companies and data center operators, and they will ask about reliability, connectivity, maintenance, insurance and maritime rules long before price. Marketing here looks less like advertising and more like technical papers, pilot results and partnerships that lend credibility.

The RFS tells you what YC's partners and a few founders find promising. It does not tell you who will pay, how they choose a vendor or what they need to see before they trust a young company. That is the marketing question, and as the 13 ideas show, the answer changes with the buyer.

• An idea from the list does not bring customers with it. The market still has to be won, one buyer at a time.
• It is not a route into YC. YC says the ideas are only a fraction of what it funds and that you do not need to work on one to apply.
• The list changes. This one is labeled Fall 2026 and presents its ideas as this batch's requests.
• Buyers can tell you what they need to see. Whatever the idea, the people who would pay for it know what evidence would convince them.

Marketing starts from that evidence and from the buyer's own words. If you are building a technology company and marketing is on your list, our page on marketing for technology companies explains how we work.