YC summer requests favor AI-native services

Y Combinator just gave founders a pretty direct signal about what it wants next. Not more AI add-ons. Not another copilot bolted onto old software. The Summer 2026 Req(ycombinator.com)nd the sharpest implication is in one category: AI-native service companies that sell the outcome, not the tool. (ycombinator.com)lways matter because founders treat them like a map of partner appetite. This Summer 2026 list widened out into software, services, (ycombinator.com)ing company is no longer the assistant that helps a human do the job. It is the company that does the job itself. (ycombinator.com) ### What does “AI-native service company” mean? YC’s wording is very clear here. Services used to become SaaS. Then they became copilots. From 2023 through 2025, that was the(ycombinator.com)s that “don’t sell software—they sell the service” and “just do the work.” That is a much bigger claim than workflow automation. It means the product is the completed task. (ycombinator.com) ### Why is that a bigger market? Because services spending dwarfs software spending. YC makes that point directly, and it matter(ycombinator.com) seats to a team, the budget line it can attack is much larger. That is why this reads less like “build a smarter app” and more like “rebuild a business category around model-driven execution.” (ycombinator.com) ### Why now? YC’s answer is basically model capability. The page argues that AI systems can now handle complex work well beyond engineering, which change(ycombinator.com)tive service starts to make sense when the model can take in requests, route work, generate outputs, and recover from edge cases with much less human glue. (ycombinator.com) ### Where does the “central brain” idea come from? The exact phrase is circulating in commentary more than on YC’s page, but the architecture is implied by the reques(ycombinator.com)sions, handoffs, and exceptions. That “something” is a model layer acting as the operating core — less like a feature inside software, more like the service’s brain. That is an inference from YC’s framing, but it fits the whole document. (ycombinator.com) ### Why are people pointing at healthcare examples? Because healthcare (ycombinator.com)orm using AI for symptom analysis, diagnostic hypotheses, and treatment suggestions across chat, video, and in-person care. Whether Kompa itself becomes the breakout winner is a separate question. The reason it gets mentioned is structural: triage is a service, demand is constant, and the user wants an answer, not software. (kompa.com.br) ### Is this just investor fashion? Partly, sure. But it is also a reaction to the last wave. Copilot startups were easier to launch, (ycombinator.com)s with deeper operational moats — distribution, workflows, compliance, and real service delivery — where AI is embedded in the product’s core economics, not sitting on top as a shiny layer. (ycombinator.com) ### What’s the catch? The catch is that selling a service is harder than shipping software. You inherit quality control, exceptions, trust, and often regulation. Healthcare makes that obvious. Research on ge(kompa.com.br) can look decent while triage still lags physicians by a meaningful margin, so “AI does the work” is not the same as “AI can be left alone.” (thelancet.com) ### Bottom line? YC’s Summer 2026 message is not “use AI.” Everybody already does. The message is to build companies where AI is the operating core and the customer buys th(ycombinator.com)ss like software vendors and more like entirely new service firms. (ycombinator.com)