Large language models enable rapid prototyping of AI via prompt programming. But while prompt programming makes it possible to prototype AI in record time, we’re still collectively defining what it means to “prompt program”: What are best practices? What types of tools are useful? What types of problems can even be solved via prompt programming? In this talk, we share our experiences in researching and developing tools to support prompt programming in an industry setting. We describe the opportunities that prompt programming creates, the challenges of defining computation in this medium, and open research problems for this new way of prototyping computation.