The real bottleneck for AI agents isn't the artificial intelligence itself; it's the application programming interfaces (APIs) that connect AI systems to actual business data and workflows. While organizations can easily access powerful AI foundation models from multiple providers, the challenge emerges when those models need to interact with real systems, business processes, and production data. According to Gartner research, over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.

Postman, a platform for designing and managing APIs, recently achieved AWS AI Competency recognition in the Agentic AI Tools category. This designation reflects a critical shift in how enterprises think about AI infrastructure. The AWS AI Competency isn't a self-certification; it requires technical validation and demonstration of successful customer implementations against standards for security, reliability, and operational excellence.

Why Do AI Agents Fail When APIs Are Poorly Designed?

AI agents depend on APIs to reach systems, retrieve information, run workflows, and take action. Unlike human developers, who can infer intent from ambiguous documentation and adapt on the fly, AI agents cannot. If an OpenAPI specification omits authentication scopes, misrepresents response schemas, or fails to document error codes and rate-limit behavior, the agent fails, either silently or loudly.

The specificity problem is fundamental. A human developer can work through unclear API documentation with effort and creativity. An agent cannot. As AI agents become a distinct buyer class that discovers services, evaluates options, and completes transactions independently, the quality, discoverability, and governance of an organization's APIs will directly determine whether that organization can participate in an agent-driven economy.

How to Build APIs That AI Agents Can Actually Use

• Design Well-Specified APIs: Create clear, comprehensive OpenAPI specifications that document authentication requirements, response schemas, error codes, and rate-limit behavior so agents can understand and call endpoints reliably.
• Enforce Design Consistency: Use API governance rules and automated validation tools to ensure all APIs follow the same patterns and standards across your organization, reducing the friction agents encounter when discovering and integrating services.
• Validate API Behavior Before Deployment: Run automated test suites that verify API responses and behavior before agents ever call an endpoint, catching issues that would cause silent failures in production.
• Structure Collections for Machine Readability: Organize API documentation, request examples, and tests in a way that is machine-readable by design, making them usable by both human developers and AI agents without a separate integration layer.

PayPal demonstrates what happens when an organization treats its APIs as strategic infrastructure built for both human developers and AI agents. Since publishing their public Postman workspace, PayPal's API collections have accumulated more than 100,000 forks, placing them among the most-forked on the Postman Public API Network.

The business impact is concrete. Time to first API call dropped from 60 minutes to 1 minute, and testing cycles that used to take hours now finish in minutes. Engineers on the Postman Enterprise plan save roughly one hour per week on API development workflows.

"Postman has become a front door to PayPal, and increasingly the developer walking through it is an AI agent. We built our collections and Flows so humans and agents read them the same way, turning discovery into a first API call in under a minute and, over three years, more than 100,000 forks," said Mark Lummus, Head of Product, Developer Tools at PayPal.

Mark Lummus, Head of Product, Developer Tools at PayPal

PayPal published its Model Context Protocol (MCP) server as a Postman Collection, giving developers a ready-to-use, fully documented set of API requests covering payments, invoices, disputes, shipment tracking, subscriptions, and more. The collection uses Postman's built-in OAuth support, which reduces the authentication friction that typically slows agent integration. The result is that an AI agent can discover, authenticate, and call PayPal's commerce APIs using the same collection infrastructure that human developers already rely on.

What Does the AWS AI Competency Actually Validate?

The AWS AI Competency in Agentic AI Tools covers platforms that support the full development lifecycle, from design and specification through testing, governance, and production operations, across different regulatory environments. This recognition reflects capabilities enterprises need to move AI from experimentation to production and to build the engineering discipline required to operate AI systems at scale.

Gartner estimates that of the thousands of vendors claiming agentic AI capabilities, only around 130 represent genuinely agentic products. The model selection problem is largely solved; the infrastructure problem is not. Getting APIs ready for how agents actually discover, understand, test, and call them reliably at scale is where most AI initiatives hit friction. Add governance, compliance, and regulatory requirements for enterprises in regulated industries, and the challenge compounds.

As AI agents become more prevalent in enterprise environments, the quality of API infrastructure will become a competitive advantage. Organizations that invest in well-designed, thoroughly tested, and properly documented APIs will be positioned to deploy AI agents at scale. Those that neglect API quality will find their AI initiatives stalled by the very infrastructure that should enable them.