企业级 Agent 部署仍不到 1%，但拐点已至Enterprise Agents Still Under 1% Deployed, But Inflection Point Arrives
Box CEO Aaron Levie 指出，绝大多数企业工作流其实还没有真正用上 agent，当前经济体中的部署率可能不到 1%。现有的最多只是单次 LLM 调用做决策或反馈，真正的自动化极少。这既有企业技术采用慢的原因，更核心是 AI 能力本身的限制。如今随着 computer use、长时运行 agent、harness 改进和周边基础设施成熟，终于可以对更复杂的工作流动手。未来会出现自改进系统，能针对任何工作流目标在大量实例上提升执行率，而 agent swarm 形态让人可以任意投放算力去解决问题。这不再是终端用户提示 agent 完成单任务，而是组织重新设计底层工作流以发挥 agent 优势，对金融、生命科学、医疗等重信息领域将产生巨大影响。
Box CEO Aaron Levie notes that the vast majority of enterprise workflows have not meaningfully adopted agents yet, with economy-wide deployment likely under 1%. At best, companies have a few systems making a single LLM pass for decisions or feedback, but almost no real automation. Part of this is slow enterprise tech adoption, but most stems from AI capability limits themselves. With computer use, long-running agents, harness improvements and supporting infrastructure now available, complex workflows are finally in reach. Self-improving systems will raise execution rates across volumes of instances for any workflow goal, while agent swarms let organizations throw arbitrary compute at problems. This shifts from end users prompting agents for one-off tasks to organizations reengineering core processes around agents, creating massive opportunity in finance, life sciences, healthcare and other information-heavy domains.
Box CEO Aaron Levie notes that the vast majority of enterprise workflows have not meaningfully adopted agents yet, with economy-wide deployment likely under 1%. At best, companies have a few systems making a single LLM pass for decisions or feedback, but almost no real automation. Part of this is slow enterprise tech adoption, but most stems from AI capability limits themselves. With computer use, long-running agents, harness improvements and supporting infrastructure now available, complex workflows are finally in reach. Self-improving systems will raise execution rates across volumes of instances for any workflow goal, while agent swarms let organizations throw arbitrary compute at problems. This shifts from end users prompting agents for one-off tasks to organizations reengineering core processes around agents, creating massive opportunity in finance, life sciences, healthcare and other information-heavy domains.
Box CEO Aaron Levie notes that the vast majority of enterprise workflows have not meaningfully adopted agents yet, with economy-wide deployment likely under 1%. At best, companies have a few systems making a single LLM pass for decisions or feedback, but almost no real automation. Part of this is slow enterprise tech adoption, but most stems from AI capability limits themselves. With computer use, long-running agents, harness improvements and supporting infrastructure now available, complex workflows are finally in reach. Self-improving systems will raise execution rates across volumes of instances for any workflow goal, while agent swarms let organizations throw arbitrary compute at problems. This shifts from end users prompting agents for one-off tasks to organizations reengineering core processes around agents, creating massive opportunity in finance, life sciences, healthcare and other information-heavy domains.
查看原文 →