The declarative approach to API orchestration, popularized by Apollo Connectors, has several fundamental technical problems. These include GraphQL Directives not being designed for infrastructure as code, the verbosity of the `@connect` directive making it hard to understand and maintain complex scenarios, the leaking of implementation details into the GraphQL SDL, and the N+1 problem caused by the declarative nature of the approach. Additionally, testing and debugging connectors is challenging due to their reliance on human-centric workflows and the lack of standard testing frameworks. In contrast, gRPC Plugins, which utilize Generative AI (LLMs) to generate code for API orchestration, offer a more scalable, efficient, and maintainable solution with features such as easy testability, high performance, and support for multiple languages.