AWS just shipped an experimental library through strands-labs, AI Functions, which execute LLM-generated code at runtime and return native Python objects. They use automated post-conditions to verify outputs continuously. Unlike generate-and-verify approaches, the AI-generated code runs directly in your application.