How Postman runs Agent Mode for 40 million developers on Amazon Bedrock
Building an AI agent that works in a demo is a different problem from running one for 40 million developers. Postman and AWS share the architectural patterns behind Agent Mode: controlling tool sprawl, exposing schema-based reads, and treating context as the real bottleneck, plus how it runs on Amazon Bedrock at scale.
Postman has developed Agent Mode, an AI-native portal for developers to interact with the platform across testing, documentation, discovery, and implementation. The team anticipated challenges in model quality and prompt design, but the most significant hurdles arose from integrating an agent into an existing, mature product with years of interface-driven assumptions, a vast array of tools, and specialized concepts.
To address these issues, Postman leverages Amazon Bedrock for model flexibility, geographically scoped cross-Region inference, model-dependent zero data retention, and multi-tier prompt caching. By using these Amazon Bedrock features, Postman can scale production workloads without managing its own model-serving infrastructure while maintaining control over model selection, throughput, geographic processing, and costs.
Agent Mode operates directly against the Postman application, opening pull requests and proposing next steps without requiring user navigation. Human oversight is an essential part of the design, as the agent requires user approval before taking any actions that modify application state. Postman also applies controls to reduce unintended actions and data exposure, such as limiting available tools based on task requirements, using purpose-built context, and configuring model-dependent data-retention settings.
To tackle tool sprawl, Postman selects tools based on specific needs and context, isolating individual execution threads so the model only sees the relevant tools for the current task. This dynamic tool selection process allows for improved effectiveness, even when the toolset size grows larger. Additionally, the team is actively working to decouple tools from interface state, enabling the agent to perform tasks without needing specific interface elements to be open.
When dealing with structured data like service uptime, test results, and endpoint response times, Agent Mode can generate complex queries using a consolidated query tool. By treating tool catalogs as part of the context budget and dynamically exposing only the necessary tools for each task, Postman aims to make production agents more effective and beneficial for their global developer community.
Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
