Google has announced significant updates to the Gemini API's Structured Outputs, focusing on enhancing support for JSON Schema and improving property ordering in outputs. These changes address the needs of developers who rely on structured data for tasks like data extraction and database population. The updates facilitate smoother communication between AI agents, allowing one agent's output to serve as another's input without needing additional translation. With the expanded JSON Schema support, developers using libraries like Pydantic for Python or Zod for JavaScript/TypeScript can integrate more easily with the Gemini API. This includes support for various JSON Schema keywords that many developers have requested, such as `anyOf` for conditional structures, `$ref` for recursive schemas, and constraints for numeric values. These enhancements aim to make it easier for developers to ensure their data follows a specified format. The API now also preserves the order of keys as defined in the schema, which is particularly useful for Gemini 2.5 models and the OpenAI compatibility API. For example, Agentic Users has leveraged Structured Outputs to streamline data extraction related to brand guidelines, resulting in significant cost savings. Similarly, Alkimi AI utilizes JSON Schema to manage the configuration of their AI assistants, ensuring reliable data flow through their multi-stage language model pipeline. These changes are effective immediately, and developers can access the updated features through the API documentation.