Postman rolls out Agent Mode and AI updates aimed at real API work
Postman introduced Agent Mode and a set of AI-focused updates that go beyond generic code assistants. The pitch: agents trained for API design, testing and development that work with your specs, test results and production signals. The goal is fewer hallucinations, tighter scope and faster iteration on the actual units that matter - your requests, collections and environments.
Postman also expanded governance and distribution features so teams can see which APIs exist, who owns them and how they behave across dev, test and production. That matters as agentic workflows spin up more surface area and more endpoints, which need to be discoverable and controlled.
What's new
- Agent Mode: Summarizes APIs, diagnoses issues and applies updates using API-specialized LLMs and agents.
- API Catalog: A consolidated view of APIs across environments with ownership and usage context for governance.
- Git workflows: Manage specs, collections, tests, mocks and environments directly in local repos.
- MCP servers: Integrations with Atlassian, Amazon CloudWatch, GitHub, Linear, Sentry and Webflow via MCP (Model Context Protocol).
- Distribution and publishing: Documentation, workflows, sandboxes and SDKs in one place for easier rollout.
Why this matters for API teams
Agentic AI is driving an explosion in internal and partner-facing APIs. That's useful for safety and least-privilege access, but it amplifies cataloging, testing and drift problems. As one analyst put it, "Context engineering is the real differentiator" - and Postman is betting it can keep the context advantage by owning the specs, tests and production behavior agents rely on.
The open question is whether that advantage holds as IDE-native assistants, observability platforms and platform engineering stacks assemble similar context from different angles. Expect faster iteration, but also stricter standards to keep sprawl and tech debt in check.
Customer perspective: quality over vibe coding
One Postman customer said specialized agents for API work are more compelling than last year's "vibe coding" hype. "API quality and consistency are more important than ever because of AI," he noted, warning that acceleration without standards creates debt that kills delivery speed.
He views Agent Mode as a step up from single-turn chatbots: "The problem with chatbots is that they are single-turn… With a multi-turn agentic system, the possibilities expand." He also called Postman's new git integration a smart fit with how teams already work alongside tools like GitHub Copilot and Claude Code.
PayPal example: natural-language API workflows in the wild
PayPal runs the most-forked public collection on Postman, with more than 100,000 forks. Agent Mode is now enabled for that collection and tied to PayPal's MCP server, so developers can ask, "Create an invoice for $100 and send it to this email," and the agent handles the flow.
That experience demands APIs that are readable by both humans and agents. PayPal highlighted the need for explicit, step-by-step guidance in docs - even simple files like an "llms.txt" that spells out how the agent should interpret instructions. Internally, PayPal uses APIMatic for API development and Postman for distribution, with a push toward a consistent mental model across frameworks and IDE tools.
Practical next steps for engineering leaders
- Map your API surface: Use a catalog to identify owners, environments and consumers. Close gaps before adding agents.
- Codify quality gates: Enforce linting, contract tests and approval workflows on every PR to avoid agent-induced sprawl.
- Adopt multi-turn agents where they add leverage: Start with high-churn tasks (spec updates, test generation, regression fixes).
- Wire in git-first workflows: Keep collections, tests and environments versioned alongside code to align with dev habits and AI assistants.
- Standardize agent-facing documentation: Publish machine-readable instructions (for example, an "llms.txt") that describe auth, pagination, rate limits and error handling.
- Use MCP to scope and secure integrations: Treat each back-end capability as a narrow, governed interface; see the Model Context Protocol for patterns.
- Instrument and learn: Feed test and production behavior back into your catalog so agents act on current reality, not stale assumptions.
For deeper skill-building
Your membership also unlocks:
