Researchers propose a new API paradigm for enterprise AI agent systems called Agent-First Tool API. The current APIs used in AI research are based on human-oriented CRUD (Create, Read, Update, Delete) paradigms and do not meet the requirements of autonomous agents. The proposed paradigm addresses five architectural mismatches between conventional APIs and agent requirements, including exact identifier dependence and opaque error semantics. It includes a Six-Verb Semantic Protocol that decomposes tool interactions into different phases, a Normalized Tool Contract providing decision-support metadata, and a dual-layer governance pipeline. The researchers implemented and validated the paradigm in a production multi-tenant SaaS platform, demonstrating improved task success rates and reduced human interventions. --- Why it matters: This matters to AI engineers because it addresses a critical issue with current APIs used in enterprise AI systems, which can lead to inefficiencies and errors. By adopting this new paradigm, developers can create more autonomous and robust AI agents that require less human intervention. Source: https://arxiv.org/abs/2605.10555