YC: Requests for Startups - 2026 Fall

Thirteen open problems, and what you could ship against each in a week

Y Combinator's Request for Startups for Fall 2026 lists 13 areas its partners want to fund. Read as a wish list it is mildly interesting. Read as market research it is more useful than that: a group of people who see thousands of companies a year, publishing where they think the gaps are.

The failure mode is treating each entry as a mandate to build the whole category. Nobody funds a category. What follows is each RFS, and next to it the smallest thing we would actually build first: something with one customer type, one workflow, and a way to charge for it. The wedges are ours, not YC's.

A request for startups is a list of markets where someone credible has already done the demand research for you.

Consumer and care

• The Primer. YC wants a tutor that adaptively teaches young children to read, write and do arithmetic at private-tutor quality. Wedge: phonics only, ages four to seven, ten minutes a day on a parent's phone, with a weekly progress note that tells the parent exactly which sound their child is stuck on. One skill ladder done properly beats a whole curriculum done shallowly, and the progress note is what actually gets renewed.

• AI-Powered Consumer Products for 1 Billion People. Consumer AI across transport, learning, health, finance and entertainment. Wedge: pick one boring category in one underserved language market. A budgeting assistant that runs entirely inside WhatsApp, reads the bank notification texts people already receive, and answers in local currency and local idiom. No app install, no onboarding flow.

• AI for the Aging Population. Voice interfaces, safety monitoring, caregiver coordination. Wedge: a daily check-in phone call to one senior, in their own language, that escalates to an adult child if something sounds wrong. No hardware, no app for the senior to learn, and the person who pays is the one worrying at 2am.

Software infrastructure

• A Cloud for Small Software. Infrastructure for small, personal tools rather than enterprise deployments. Wedge: one-click hosting for the single-purpose apps agents now generate constantly, with auth, a database and a share link included. Priced per app per month. The market is every internal tool currently dying on somebody's laptop.

• Multiplayer AI. Teams dropping into the same live agent session to observe and redirect. Wedge: shared sessions with a URL, presence indicators and takeover, wrapped around one existing coding agent, sold to teams of three to ten. Essentially collaborative editing for agent runs, which is a feature before it is a platform.

• Self-Maintaining APIs. Agents that update customer code when a provider ships breaking changes. Wedge: subscribe to exactly one high-churn API's changelog, watch customer repositories, and open a tested pull request the day a breaking change lands. Charge per repository. Expand to a second API only once the first one runs unattended.

Trust and compliance

• Proving You're Human. A verified-human layer for communication and transactions, without surrendering privacy. Wedge: payment-change verification for finance teams. Before any bank detail change is actioned, the system runs a verified challenge against the requester. Deepfaked executive requests are already costing companies real money, which makes this a budget line rather than a nice-to-have.

• AI-Native Compliance Infrastructure. Consolidated, real-time compliance instead of fragmented manual tooling. Wedge: one framework, one company stage. Evidence collection for a single certification at seed-stage SaaS companies, where the agent watches the codebase and infrastructure and keeps the evidence current instead of producing it in a panic each audit.

The physical world

• New Operating Systems for the Physical World. Software coordinating humans, robots and agents in construction, maintenance and logistics. Wedge: dispatch for one trade in one city. Jobs arrive by phone and email, an agent schedules them and confirms with the customer, and the operator sees one board. Start where the incumbent tool is a spreadsheet.

• Data for the Real World. Dense physical-world data collection to model energy, agriculture, logistics and construction. Wedge: instrument one narrow thing well. Cold chain monitoring for food distribution, sold per vehicle per month, where the value is a temperature excursion alert that prevents a rejected delivery.

• The Future of American Defense. Interceptors, sensors, drones, resilient logistics and advanced manufacturing. Wedge: be honest that hardware here is neither simple nor fast, and the procurement cycle will outlast your runway. The tractable entry is software for the suppliers: tooling that helps small manufacturers navigate export control and bid preparation. It is unglamorous and it is where a small team can actually start.

• Compute at Sea. Offshore data centre flotillas exploiting free cooling and no permitting. Wedge: there isn't a simple one, and pretending otherwise would be dishonest. This is a capital and maritime engineering problem. If you are drawn to it without a shipyard, the adjacent software business is siting and thermal modelling for operators who do have one.

Money

• The Best Time to Build in Crypto. Stablecoins, tokenised assets, agentic commerce, emerging market rails. Wedge: stablecoin payouts for cross-border contractors in a single corridor, bundled with the invoicing and tax paperwork nobody wants to do. The crypto part is plumbing the customer should never have to think about.

What this list actually tells you

Eleven of these thirteen have a first version that a small competent team could put in front of a paying customer within weeks. That was not true three years ago, and it is the part of the current moment that gets underdiscussed. The constraint moved off building.

Which is where our work sits. We build the first shippable version of things like these: scoped down to one workflow, run through parallel agent sessions inside isolated environments, reviewed adversarially so that speed does not quietly become debt, with a person accountable for every merge. The framework routes each task to the right model rather than paying frontier prices for work a cheaper fine-tuned model does equally well, which is what keeps a fast build from becoming an expensive one.

The two entries we flagged as not simple are the interesting test. An honest partner tells you which of your ideas is a software problem and which is a shipyard problem, before you spend a year discovering it. That judgement has not been automated either.