We’re seeing the shift: AI agents are taking over task execution, relying entirely on the APIs we build. If a human engineer can squint at a 400 Bad Request and figure out the JSON structure, an Agent needs a machine-readable, flawless contract (OpenAPI spec, great docs, predictable errors).
Is this the moment we stop prioritizing new feature volume and finally invest in “boring” API quality, governance, and OpenAPI/AsyncAPI specification maintenance? What’s the biggest threat to Agent-based workflows: The model’s intelligence, or the state of your 10-year-old API documentation?
Look, we’ve heard this before. “Microservices will enforce perfect contract testing!” “Serverless will kill boilerplate!” What happened? Humans wrote code and made deadlines. The biggest threat is still an engineering manager demanding a feature that bypasses all the “boring” governance. Good docs? Sure. Perfect? Never.
This is the forcing function we needed. We want Agents using our APIs; that’s leverage. The need for precise tooling (like HAR file generation and better LLM usage tracking, per the InfoQ news) proves it. Companies that invest in self-describing, machine-readable contracts now will win the A2A (Agent-to-Agent) economy. It’s an easy, justifiable decision.
It’s going to be a shadow IT nightmare, part two. Instead of building new features, we’re just going to glue Agents to our existing garbage APIs and call it ‘AI-driven business transformation.’ We won’t fix the API; we’ll just build a flaky ‘Agent Adapter’ layer that costs twice as much to maintain. You heard it here first.
Spot on. The Agent is the ultimate consumer-of-API. It has no tolerance for undocumented side effects or mutable state without a clear transactional boundary. This isn’t about documentation; it’s about architectural rigor. We need to move back to pure governance-first backend design where the API contract is the source of truth, not a descriptive afterthought.