Topics for Y Combinator – Fall 2026
The Y Combinator has posted the topics (products) it wants to cover and support.
Here is a list of them:
The Primer – AI education with personalised learning
The Future of American Defense (Drones, sensors, interceptors,...)
A Cloud for Small Software
Multiplayer AI (Teams collaborating with shared AI agents)
Compute at Sea (Offshore data centres)
AI-Powered Consumer Products for 1 Billion People
AI for the Ageing Population
New Operating Systems for the Physical World
The Best Time to Build in Crypto (stablecoins, agentic commerce)
Data for the Real World
Proving You're Human
AI-Native Compliance Infrastructure
Self-Maintaining APIs
Which of these do you find most interesting, and which ones are missing from the list?
As for me, I like the topics: AI for the Ageing Population and Proving You're Human.
But what I am really missing is AI for health care and longevity.
Replies
@busmark_w_nika I really like AI for the Ageing Population from this list. And of course, Proving You’re Human - maybe the big thing we all need today =)
Submitted an application:
AI-Powered Consumer Products for 1 Billion People. Any chances? I guess not ) But we need to try
@konstantin_tikhaev I am curious how this will unfold. Fingers crossed :)
AI-Powered Consumer Products for 1 Billion People is the one I'd bet on, consumer AI is still mostly chat wrappers and the tools people use daily haven't been rebuilt yet. What I'm missing on that list is anything about trust: the products that win the next billion will be the ones that can prove where your data goes, not just promise it. Proving You're Human is close, but it solves that for platforms, not for people.
Proving You're Human is the one I'd bet on and also the one I think gets built wrong first. Nearly every version I've seen verifies a human was there at signup, which isn't the same as a human being there for the action that matters. We saw a small version of that on free credits at Zeplik, and nothing at the signup gate helped, the only useful signal showed up mid session. What I'd add to the list is the unglamorous other half of Multiplayer AI, the record of what a shared agent actually did once it was allowed to act. Everyone is building the permission and nobody is building the receipt.
@asadmalik901 Lately, those "approval" processes for humans are so difficult even for humans :D
@busmark_w_nika Ha, and that's the whole problem in one line. We keep making the test harder for people while the bots quietly get better at passing it. The version that works probably won't feel like a test at all, it'll just watch what you do during the session and stay quiet, which is a much harder thing to sell than a checkbox.
@asadmalik901 Maybe next thing should be: If you do not complete the test, you are 100% human :)
@busmark_w_nika Not far off actually. The strongest signal we had wasn't a test, it was hesitation. Real people pause before they click and bots don't, and that gap beat anything we ever put on the signup form. Problem is you can't put "we noticed your mouse was indecisive" on a pricing page.
Self-maintaining API's really caught my eye here. That's a real problem out there that I am currently facing with my build, so definitely a lot of potential there.
@aamirxv2 Are you building something related?
Proving You're Human is going to be huuuuge 😁
Self maintaining apis is the one id actually build off that list, because it gets more valuable as everything else on the list ships. every agent product up there breaks the moment an upstream endpoint quietly changes shape. whats missing for me sits right next to proving youre human. that one assumes the problem is identity, and the harder version is proving an action was authorised. an authenticated agent doing something its owner never intended passes every human check on that list, and nobody is building the layer that decides who pays when it does. ageing population is the one i think is commercially underrated, because the buyer and the user are different people, and that is usually where the pricing power sits.