Agent-First Tool API: A Semantic Interface Paradigm for Enterprise AI Agent Systems
A new research paper published on arXiv introduces the 'Agent-First Tool API,' a semantic interface paradigm designed to bridge the gap between conventional human-oriented APIs and the needs of autonomous enterprise AI agents. The study identifies five architectural mismatches in current CRUD-based interfaces, such as exact-identifier dependence and opaque error semantics. To address these, the authors propose a framework featuring a Six-Verb Semantic Protocol (search, resolve, preview, execute, verify, recover), a Normalized Tool Contract with decision-support metadata, and a dual-layer governance pipeline. Validated in a production multi-tenant SaaS platform across six business domains, the paradigm demonstrated an 88% end-to-end task success rate, significantly outperforming optimized CRUD baselines by 37.5%. Additionally, it reduced human intervention requirements by 72.7% and improved autonomous error recovery by 5.8 times. The research establishes this paradigm as a complementary semantic application layer operating above existing transport-layer standards like MCP, marking a significant advancement in optimizing tool interactions for scalable enterprise AI systems.
Wire timeline
Agent-First Tool API: A Semantic Interface Paradigm for Enterprise AI Agent Systems
A new research paper published on arXiv introduces the 'Agent-First Tool API,' a semantic interface paradigm designed to bridge the gap between conventional human-oriented APIs and the needs of autonomous enterprise AI agents. The study identifies five architectural mismatches in current CRUD-based interfaces, such as exact-identifier dependence and opaque error semantics. To address these, the authors propose a framework featuring a Six-Verb Semantic Protocol (search, resolve, preview, execute, verify, recover), a Normalized Tool Contract with decision-support metadata, and a dual-layer governance pipeline. Validated in a production multi-tenant SaaS platform across six business domains, the paradigm demonstrated an 88% end-to-end task success rate, significantly outperforming optimized CRUD baselines by 37.5%. Additionally, it reduced human intervention requirements by 72.7% and improved autonomous error recovery by 5.8 times. The research establishes this paradigm as a complementary semantic application layer operating above existing transport-layer standards like MCP, marking a significant advancement in optimizing tool interactions for scalable enterprise AI systems.
cs.AI updates on arXiv.org