企业智能体的下半场，如何让智能体越用越聪明？ (Enterprise Intelligent Agents' Second Half: How to Make Agents Smarter with Use?)

This article examines enterprise AI agent evolution and introduces Alibaba Cloud’s AgentLoop platform as a continuous-improvement solution. It identifies four enterprise difficulties: data collection complexity from agent topology structures, dataset construction challenges, multi-layer evaluation across step/trajectory/outcome levels, and asset standardization for memory, skills, and workflows. AgentLoop addresses these through trajectory observability via LoongSuite, an Agent Ontology, automated dataset pipelines, Agent-as-a-Judge evaluation, and memory/experience repositories that feed successful patterns back into agent contexts.