Google shares a design pattern for AI agents to fetch and validate schemas at runtime instead of relying on static prompts.
Google has outlined a design pattern for building AI agents that fetch and validate data schemas at runtime instead of loading all rules into the initial prompt.
The approach, called Context-Aware Polymorphic Schema Validation, is designed for developers building multi-agent workflows with Google’s Agent Development Kit and Gemini Flash.
Google said static prompt-based systems can become harder to manage as workflows expand. When developers preload every schema, tool definition and validation rule, prompts can become larger, slower and more prone to confusion between similar data structures.
Under the proposed model, schemas are stored in a central metadata registry, such as Cloud Storage. The agent first identifies the user’s intent, then retrieves only the relevant schema for that task.
Each field is validated through separate programmatic checks, such as format, range or pattern validation, before being added to the final structured payload.
Google said this separation can reduce token use, improve reliability and make it easier to update business rules without redeploying application code.
The design is also intended to improve handoffs between agents by ensuring data is checked before moving into downstream enterprise applications or another agent workflow.
The broader message is that scalable AI agent systems may need to separate reasoning from structural validation, allowing models to focus on context while external systems enforce data accuracy.
