The problem showed up at PayPal DevDay 2025: developers watching AI agents integrate with PayPal's APIs noticed the agents kept grabbing old documentation and deprecated SDK versions. Instead of reading current API specs, the agents defaulted to whatever they'd absorbed during training - which, depending on the API, could be months or years out of date.
Sohaib and Hannan from APIMatic saw the same pattern. Their company has spent five years building tools to help human developers work with APIs. Now they're applying that same expertise to AI coding agents, which face a structurally different version of the same problem. A human developer can search for the latest SDK version. An AI agent, left to its own devices, falls back on its training data.
Their answer is context plugins - structured packages of current API documentation that AI coding agents (like Cursor, Claude Code, or Amazon Q Developer) can pull from directly instead of guessing from memory. The idea is simple: anchor the agent to today's actual specs rather than hoping its training data is current.
For fast-moving APIs, this matters more than it might seem. Stripe, Twilio, and similar platforms push frequent updates. An agent writing integration code from its training snapshot might call endpoints that no longer exist or pass parameters that were deprecated two versions back. The resulting code compiles fine and fails at runtime - the worst kind of bug to debug.
APIMatic has posted a showcase on their site but hasn't announced pricing or general availability. This is a product preview, not a full release. The underlying problem they're targeting is real and widespread - most developers using AI coding tools for API integrations have hit exactly this failure mode at least once. Whether APIMatic's approach becomes the standard fix, or whether the major AI coding tool vendors solve it from the agent side, is an open question.