Integrating with services often leads to frustration due to unreliable documentation. Developers frequently encounter issues like outdated examples and missing details that can result in errors like "400 Bad Request" or "422 Unprocessable Entity." The author argues that the solution isn't to improve documentation but to shift to using types as the primary contract between services. As AI agents increasingly consume APIs, they benefit from direct type definitions, enabling them to generate valid code on the first attempt without relying on potentially outdated documentation. The article emphasizes that documentation is a "lossy copy" of the actual code. It can easily fall out of sync as teams prioritize feature development and bug fixes over keeping documentation updated. The author suggests that instead of writing prose to describe contracts between services, developers should focus on creating executable type definitions. These types serve as the contract, ensuring that both the producer and consumer adhere to the same standards without the risk of drift that comes with manual updates. Using JavaScript-native RPC, developers can invoke methods between services with type safety derived from schemas, eliminating the guessing game that often accompanies traditional HTTP APIs. The author illustrates this with a practical example that shows how defining a schema results in consistent types across services, frontend applications, and documentation. This approach allows for seamless integration and reduces the likelihood of runtime errors, making the codebase more efficient and reliable. The integration of AI agents into this system allows them to work more effectively, reducing latency and error rates significantly when they can leverage type definitions directly.