Stories about 李凯
1 related stories
美团企业级 Agent 的平台化应用实践｜QCon上海
AI InsightMeituan's enterprise-grade Agent evolution from 'running' to 'trustworthy' signifies that the core of LLM adoption has shifted from model capability competition to engineering infrastructure and task collaboration systems. By systematically isolating and governing execution, permissions, and state via AgentOS and Multica, enterprises are converting uncontrollable unstructured model outputs into production-ready, supervisable deterministic services.Key TakeawayEnterprise-grade Agents are shifting from isolated demo validations to platform-based production centered on AgentOS and engineering infrastructure.Why It MattersModel output uncertainty and long-chain state management are primary roadblocks to enterprise AI adoption. Establishing a complete engineering system including sandboxes, permission governance, state management, and human-agent collaboration determines whether AI apps can scale past cost and security constraints.Who's Affected- BeneficiaryEnterprise AI DevelopersMeituan's platform engineering practices provide reusable methodologies for solving long-chain agent state and permission governance.
- WatchingAI Infrastructure ProvidersThe rise of AgentOS, sandbox management, and task collaboration frameworks is defining standards for enterprise-grade AI middleware.
What's NextFuture observations should focus on quantitative data regarding throughput performance, fault-tolerance mechanisms, and cost governance of multi-agent collaboration systems in Meituan's production environments.Importance 80/100