Postman details how it runs AI agent mode for 40 million developers using Amazon Bedrock

Postman describes Agent Mode as an AI-native interface for working across API testing, documentation, discovery, and implementation, designed to replace traditional UI navigation with agent-driven reasoning over data.

The team initially expected model quality and prompt design to be the hardest challenges, but found deeper issues in integrating an agent into a mature product with years of interface-driven assumptions and a wide surface area.

To make the product legible to an agent, Postman and AWS adopted architectural patterns including controlling tool sprawl, exposing schema-based reads, and treating context—not capability—as the primary bottleneck.

Agent Mode runs on Amazon Bedrock, which provides managed access to foundation models, enabling Postman to scale production workloads without operating its own model-serving infrastructure.

The architecture supports geographically scoped cross-Region inference, model-dependent zero data retention, and multi-tier prompt caching to meet variable, latency-sensitive demand and sharp traffic bursts from a global developer community.

Production design includes human oversight: Agent Mode requires user approval before actions that modify application state, scopes available tools to each task, and selects purpose-built context to reduce unintended actions and unnecessary data exposure.