AI-Enhanced API Design

Alexander Polomodov and Pavel Karavashkin review AI-Enhanced API Design, continuing the discussion of API governance. They compare Google with T-API at T-Bank: scale and centralization differ, yet both organizations want contract design to be more consistent and secure.

OpenAPI compatibility remains while company extensions make the contract stricter. The study uses three pillars: a controlled experiment, qualitative interviews with consumers, and expert evaluation of the result. Surveys, interviews, and telemetry reveal not only the final YAML file but also the path taken to produce it.

The experiment splits participants into three groups: standards with a linter, standards alone, and the usual process. Engineers had two hours to design an API, record their screens, and complete a survey; contract fragments were later compared for quality. Codified rules with rapid feedback helped, while documentation alone did not.

Low participation and an artificial task limit how far the numbers can travel. The conclusion is broader than one tool: work should be decomposed into observable actions, and AI applied where context and a verifiable result exist. Automated security review is promising for T-API, but people retain responsibility for the contract.