As AI agents increasingly consume APIs, specifications must evolve from human-friendly documentation to machine-optimized contracts. Amadeus engineers Anna Tsolakou and Hongliu Cao share how they built an AI-powered chatbot to maintain and update API specs at scale, addressing challenges like token bloat, context window limits, and LLM performance degradation on long inputs. Their approach uses structural splitting and chained summarizations to modify individual spec sections rather than entire files. Beyond maintenance, the broader shift requires API designers to eliminate ambiguity, enforce consistency, and write with machine readability in mind — since LLMs interpret specs literally and cannot infer undocumented constraints the way human developers can. Tools like LintGPT, Theneo, and Mintlify are emerging to help automate spec creation and enforcement.
