How to Use YC's Requests for Startups to Find Your Next Idea

If you're looking for a startup idea, stop blank-page brainstorming. The best ideas don't come from sitting in a room thinking. They come from understanding a problem so well that the solution becomes obvious. YC's latest Requests for Startups gives you ten areas to start with.

What's on the Spring 2026 List

1. Cursor for Product Managers. AI tools that help teams decide what to build, not just how.
2. AI-Native Hedge Funds. Using AI agents to develop entirely new trading strategies.
3. AI-Native Agencies. Service businesses rebuilt with software economics.
4. Stablecoin Financial Services. Financial products in the regulatory space between DeFi and TradFi.
5. AI for Government. Helping government automate manual forms processing.
6. Modern Metal Mills. Software-defined manufacturing to cut lead times.
7. AI Guidance for Physical Work. Real-time AI assistance for field workers via smart devices.
8. Large Spatial Models. Foundation models for 3D and spatial reasoning.
9. Infra for Government Fraud Hunters. Tools to accelerate whistleblower law firms and fraud investigators.
10. Make LLMs Easy to Train. Better tooling for model training and specialization.

But don't rush to pick one just yet. Read to the end first. The full list is at ycombinator.com/rfs.

The Pattern Worth Noticing

Read the list again and a clear pattern emerges. Most items are about applying AI to existing, often unglamorous industries like metal mills or government offices.

YC is telling you that the model layer is commoditizing fast. The next big companies will be built by people who understand a specific domain and know how to rewire it using the tools that already exist. If you know an industry from the inside, you have an advantage that no amount of technical skill can replace.

How to Actually Use This List

If you're a founder looking for ideas, or someone thinking about becoming one, here's how I'd use the RFS:

Pick one item that overlaps with something you already know. If you've worked in government, look at items 5 and 9. If you've managed a service business, look at item 3. If you're technical and interested in robotics, look at items 7 and 8. Every business starts with a real problem that real people experience. Your job is to find one you actually understand.

Talk to people and name the problem. Spend a week in the space. Talk to practitioners. Read what they complain about. Then try to describe the pain in one clear sentence from their perspective. Not "AI for government" but "case workers spend 6 hours a day copy-pasting between three systems." If you can't state the problem clearly, you don't understand it yet.

See if a solution starts to form. It doesn't have to be sharp. Even a blurry sense of direction is enough. The goal isn't a prototype. The goal is a problem statement that feels true and a solution you can't stop thinking about. That combination is the real signal that you've found your idea.

Before you get excited, answer three questions:

• Who is the buyer? Who writes the check? Not "government agencies" but "the deputy CIO who owns the modernization budget."
• How big is the pain or problem? Hours wasted, dollars lost, people needed.
• What forces a buyer to act now? An audit, a new regulation, a contract renewal. No trigger, no urgency, no sale.

Why I'm Personally Excited About Physical AI

Over the past year, the signals have been impossible to ignore. Tesla is developing humanoid robots. Boston Dynamics is moving toward commercial deployments. 1X and other startups are raising massive rounds (e.g., Figure). Robotics is shaping up to be the second ChatGPT moment.

The reason is simple. Multimodal models are getting more capable. And companies like NVIDIA, Hugging Face, Google DeepMind, and Physical Intelligence are making it dramatically easier to collect, standardize, and share training data for physical tasks.

If you're curious about physical AI, this is the time to start exploring.

The Bottom Line

Go pick one area from the list above. Spend a week researching. See what happens.