10 People + AI = Billion Dollar Company?

Diana says current tools can handle smaller developer tasks, such as fixing an HTML tag or a small bug, but not complex backend systems. Jared points to SWE-bench, a dataset of real GitHub issues, as a turning point for measuring progress. The discussion compares it with ImageNet, which helped drive advances in image recognition, and notes that benchmark performance alone does not show whether AI can build a product from scratch.

The speakers distinguish tidy, well-specified problems from the messier work of modeling real-world needs. They discuss how building software can help people think through what a product should do, and argue that AI may assist with implementation while people still contribute design and product judgment. Jared adds that learning to code can build logical thinking, even if AI eventually handles more programming tasks.

The panel considers whether AI could help a small team build a billion-dollar company by taking on some junior engineering work. They note that founders have long valued lean teams, while examples such as Patrick Collison show how technical founders may learn to treat company building and management as problems to solve. The speakers also question the idea that a company should be a family, suggesting that shared goals and performance may be a more useful model.

The discussion uses Jevons paradox to explain why greater efficiency does not always reduce employment: cheaper tools can increase demand. The speakers cite spreadsheets and word processors as examples, and say that easier software development has coincided with more YC applications rather than a clear trend toward smaller companies. They hope AI will help more founders turn ideas into prototypes, find early users, and attract people and capital to grow.

The speakers conclude that coding remains useful for building foundational knowledge and product judgment, even as AI takes on more routine tasks. Gary argues that easier prototyping can let more people start companies and solve customer problems. The panel expects AI to accelerate the growth in new companies, with teams of different sizes.