The People Who Built Airbnb and Stripe Just Named 7 Things They Want to Fund

OPUS CLUB · Business Insights · Published 2026-07-02

Key takeaway: YC's RFS (Requests for Startups) is not a trend report. It's a list where the partners who backed Airbnb, Stripe and Coinbase put their own names on the line and say, plainly, "we want to fund this." The Summer 2026 RFS runs to 15 categories — the most ever. YC's own framing: "AI has stopped being a feature and become a foundation." This piece walks through the seven that matter most, based on the original text.

Every season, YC publishes an RFS (Requests for Startups): a list where partners put their own names on the line and say, "start a company in this space and we'll fund you." It isn't a trend report. It's a statement of investment intent.

Below are the seven entries that matter most.

AI has stopped being a feature and become a foundation. We're excited about a new wave of startups rebuilding software, services and silicon, and pushing AI into the physical world.

YC RFS Summer 2026, introduction

📌 The Summer 2026 RFS spans 15 categories in total, including eight hardware- and capital-intensive areas such as agricultural robotics, drone defense, space electronics and lunar manufacturing. This piece focuses on the seven in software and services, based on the original text. The full list is at ycombinator.com/rfs.

01. AI-Native Service Companies

핵심: Don't build an AI accounting tool — build an AI accounting firm. Not software you sell, but a company that does the work itself.

by Gustaf Alströmer · early Airbnb Growth team → YC partner

Gustaf's logic is simple. Global spend on services dwarfs spend on software. And much of that service work is already outsourced, which makes it far easier to replace with an AI-native product.

What YC wants now is the step after the "AI copilot": companies that sell the work rather than the software. Not an "AI accounting tool" — an "AI accounting firm."

Examples include the following.

• Insurance brokerage
• Accounting, tax and audit
• Compliance
• Healthcare administration

What we find interesting now is the next step: AI-native companies that sell the service itself rather than software. Instead of handing you a tool, they just do the work.

Gustaf Alströmer, YC partner

02. Software for Agents

핵심: Most users of the internet are about to be AI agents, not people. Today's agents run awkwardly on top of software designed for humans. From here, entire software categories need to be redesigned with agents as first-class citizens.

by Aaron Epstein · founder of Colony → YC partner

AI agents can already browse the web, make purchases and manage a CRM. The problem is that they do all of it through interfaces and software built for humans. Naturally, it's slow, inconsistent, and breaks often.

Agents need an entirely different foundation: machine-readable interfaces like APIs, MCP and CLIs instead of visual ones built from forms, buttons and dashboards. They need documentation thorough enough that an agent can discover a new tool, sign up, and start using it without a human stepping in.

When everyone is building agents, the biggest opportunity may be building the software those agents depend on.

Aaron Epstein, YC partner

03. Inference Chips for Agent Workflows

핵심: GPUs were designed for "question in, answer out." Agents don't work that way. They reason across dozens of steps, double back, and hold context the whole time. That's why chips designed for agents from the ground up are needed.

by Diana Hu · YC partner

Most AI chips were designed for prompt-in, response-out inference. Agents don't behave like that. They call tools, branch, backtrack and maintain context across dozens of steps. It's a fundamentally different hardware problem.

Run an agent on today's GPUs and you get only 30–40% of peak performance, because the workload keeps swinging between memory-bound model calls, I/O-bound tool use, and CPU-bound orchestration. The other 60–70% is wasted.

That's exactly the shift NVIDIA saw coming when it acquired Groq for $20 billion.

Groq's real insight wasn't the chip. It was the compiler that made the chip work. I think the same will be true for whoever builds the next one.

Diana Hu, YC partner

04. AI for Low-Pesticide Agriculture

핵심: Instead of spraying an entire field, AI identifies each weed and pest and a robot treats only that spot. Pesticide use drops by 90% and yields go up. Garry Tan calls this a "generation-defining company."

by Garry Tan · founder of Posterous → CEO of YC

Modern agriculture runs on chemicals. As weeds and pests adapt, farmers spray more. Costs rise and margins shrink.

Three things changed at once. AI can now identify individual weeds and pests in real time. Sensors and cameras got cheap enough to deploy anywhere. And robots can act with precision. Instead of spraying a whole field, you treat a single plant.

A company that helps farmers grow more food while cutting pesticide use by 90%? That's not just a good business. That's a generation-defining company.

Garry Tan, CEO of YC

05. AI-Native Discovery Engines

핵심: Finding a new drug normally takes years: form a hypothesis, run experiments, analyze, repeat. AI can now run that entire loop on its own — not as a tool that helps researchers, but as an engine that does the discovering.

by Jon Xu · YC partner, formerly on the Amplitude product team

For centuries, scientific discovery has run on the same loop: hypothesize → experiment → interpret → repeat. The loop works, but it's slow, and every step demands significant human effort.

Frontier models have reached PhD-level performance on many scientific reasoning benchmarks. In drug discovery, materials science and protein engineering, intelligent systems are already starting to run the full design-build-test-analyze loop.

The companies that contribute meaningfully to scientific progress won't simply sell research copilots. They'll be AI-native discovery engines that propose and test hypotheses alongside researchers.

Jon Xu, YC partner

06. Dynamic Software Interfaces

핵심: Software has always given everyone the same screen. Going forward, the same app will look different for every user, because an agent assembles the interface around them.

by Ankit Gupta · YC partner

Before AI, every user of a piece of software interacted with the same interface. The only exception was enterprise software, where forward-deployed engineers designed a bespoke version for each customer. It cost a fortune, so only large enterprises got it.

YC's bet: coding agents are now good enough that every user can be their own forward-deployed engineer. Two people's email clients could look completely different — one like a task list, one like an event calendar.

Going forward, software companies will ship shared primitives, designed from the start with the expectation that users will heavily modify the final interface.

Ankit Gupta, YC partner

07. Startups That Want to Sell to Huge Companies

핵심: Selling to large enterprises used to be brutal for startups: hard to get access, products weren't deep enough, and above all, big companies wouldn't take the risk. All three have flipped. This is the first time YC has made it an official category.

by Harshita Arora + Brad Flora · YC partners

Paul Graham's long-standing advice was "startups should sell to startups," because large enterprises decided slowly and feared risk. AI broke that formula. Three enormous barriers fell at once.

Over the past three years, companies have started closing multimillion-dollar deals during the YC batch itself or within their first year — something that simply didn't happen before. Summer 2026 is the first time YC has put "sell to huge companies" in the RFS as an official category.

The single thesis running through the Summer 2026 RFS

핵심: Software is no longer a moat; it's a substrate. The $1B+ outcomes of the next decade will come from companies applying AI to physical, regulated, capital-intensive industries that legacy SaaS never touched.

References

• Official YC RFS page — ycombinator.com/rfs (Summer 2026)
• Original entries from partners Gustaf Alströmer, Aaron Epstein, Diana Hu, Garry Tan, Jon Xu, Ankit Gupta, Harshita Arora and Brad Flora