Tool contracts for AI agents: Versioned, validated, and governed tool access

Sachin Gupta demonstrates a contract-based approach to governing AI agent tool use. Agents invoke tools only by referencing a versioned contract that defines the tool action, required scopes, and input schema. A Spring Boot gateway fetches the contract, validates claims and payload shape, enforces limits, and records audit evidence before forwarding the call. The Kubernetes demo shows a valid metrics call, a denied secrets attempt due to contract mismatch, a version drift failure, and a schema validation failure with actionable feedback. Attendees leave with a contract format, a minimal reference architecture, and rollout steps.