AI is poised to significantly impact code generation, especially in distributed systems, where many bugs tend to arise. The author argues that AI should focus on programming languages and frameworks designed to minimize the expression of these bugs. Rather than generating ordinary imperative code—which would simply speed up the emergence of common distributed bugs—AI could target frameworks like Hydro, which inherently limit the types of bugs that can occur. Hydro enforces explicit contracts for distributed behavior, such as ordering and delivery, directly in the code, which helps prevent many potential issues. The article emphasizes that while testing tools like bounded model checkers (TLA+, Alloy) have made strides in exploring system behaviors, they cannot guarantee correctness. These tools only examine a finite slice of what could happen in a production environment, leaving vast areas untested. The real problem lies in the implicit assumptions that exist between components in a distributed system, which can lead to failures that are difficult to diagnose. Hydro aims to address this by making assumptions explicit at the code level, allowing for better machine checking of distributed system behaviors. The author makes a clear distinction between improving testing and improving language constructs. While traditional imperative code and model checking remain relevant, they argue that relying solely on these methods leads to debugging challenges that could be avoided with a better-designed target for AI code generation. By focusing on frameworks that enforce explicit contracts, AI can create code that is more robust and easier to reason about, reducing the number of elusive bugs that tend to emerge in complex distributed systems.