Heroku

The talk explores whether AI will let tiny teams build billion-dollar companies and whether people should still learn to code. The speakers compare coding benchmarks such as SWE-bench with ImageNet, noting that benchmarks can accelerate progress but that fixing small bugs is far from designing and building a complex product. They argue that AI can automate some programming tasks, while product design, data modeling, and understanding messy real-world needs still call for human judgment and craftsmanship. They conclude that coding remains valuable for building technical taste and problem-solving skills, and that AI may help more people turn ideas into prototypes and companies.