[
  {
    "i": 0,
    "speaker": "Speaker 1",
    "text": "欢迎各位来到Code with Claude。我是Jess，Claude Managed Agents的产品负责人。"
  },
  {
    "i": 1,
    "speaker": "Speaker 2",
    "text": "我是Michael，Claude Managed Agents的工程负责人。"
  },
  {
    "i": 2,
    "speaker": "Speaker 1",
    "text": "看到Claude Managed Agents的自然增长以及我们如何显著加速开发者的工作流程，令人非常兴奋。"
  },
  {
    "i": 3,
    "speaker": "Speaker 1",
    "text": "今天，我们想与大家分享这个故事。"
  },
  {
    "i": 4,
    "speaker": "Speaker 1",
    "text": "我们已经看到从颠覆性初创公司到最大企业的广泛采用，我们希望确保你们都能从他们所获得的经验中受益。"
  },
  {
    "i": 5,
    "speaker": "Speaker 1",
    "text": "首先，我们将探讨我们在AI能力方面看到的指数级增长，以及这对构建智能体意味着什么。"
  },
  {
    "i": 6,
    "speaker": "Speaker 1",
    "text": "我们将讨论我们在智能体开发中看到的模式，以及这些模式如何促使我们构建Claude Managed Agents。"
  },
  {
    "i": 7,
    "speaker": "Speaker 1",
    "text": "我们将介绍构建智能体的构建模块，包括其中一些最重要的部分。"
  },
  {
    "i": 8,
    "speaker": "Speaker 1",
    "text": "我们将介绍一些我们最近宣布的功能。"
  },
  {
    "i": 9,
    "speaker": "Speaker 1",
    "text": "最后，我们将与我们的英雄日本用户Rakuten进行一次非正式的对话。"
  },
  {
    "i": 10,
    "speaker": "Speaker 1",
    "text": "正如我们所看到的，模型的能力呈指数级增长，我们的期望也随之增长。"
  },
  {
    "i": 11,
    "speaker": "Speaker 1",
    "text": "我们的模型能力越先进，我们委托的工作就越复杂。"
  },
  {
    "i": 12,
    "speaker": "Speaker 1",
    "text": "我们现在看到的瓶颈越来越是基础设施，而不是实际的智能。"
  },
  {
    "i": 13,
    "speaker": "Speaker 1",
    "text": "那么，让我们把这个概念具体化。"
  },
  {
    "i": 14,
    "speaker": "Speaker 1",
    "text": "几年前，当Opus 3发布时，你可能让它编写和测试一个组件。"
  },
  {
    "i": 15,
    "speaker": "Speaker 1",
    "text": "这可能需要几分钟的专注工作。"
  },
  {
    "i": 16,
    "speaker": "Speaker 1",
    "text": "去年，当我们的Claude 4模型发布时，我们实现了升级，你可以调试一组完整的文件。"
  },
  {
    "i": 17,
    "speaker": "Speaker 1",
    "text": "你可能需要工作一两个小时，但你会在过程中进行大量指导。"
  },
  {
    "i": 18,
    "speaker": "Speaker 1",
    "text": "今年，随着我们最新的模型，你现在看到它们可以在智能体团队中整夜运行，听取你的线性待办事项列表，并在你醒来之前完成所有操作。"
  },
  {
    "i": 19,
    "speaker": "Speaker 1",
    "text": "现在，我们预计在短期内，随着像Fable这样强大的模型出现，我们将进入一个智能体能力足够完成以前需要整个团队用季度时间才能完成的任务的世界。"
  },
  {
    "i": 20,
    "speaker": "Speaker 1",
    "text": "并且智能体会完全自主地运行这个过程。"
  },
  {
    "i": 21,
    "speaker": "Speaker 1",
    "text": "因此，你可能会看到多智能体系统能够协调并全程运行整个并购流程，所需时间仅为我们现在所需时间的一小部分。"
  },
  {
    "i": 22,
    "speaker": "Speaker 1",
    "text": "因此，随着任务的逐步推进"
  },
  {
    "i": 23,
    "speaker": "Speaker 1",
    "text": "从低级指令转向端到端的结果描述。"
  },
  {
    "i": 24,
    "speaker": "Speaker 1",
    "text": "我们现在需要的远不止是提示和工具循环。"
  },
  {
    "i": 25,
    "speaker": "Speaker 1",
    "text": "我们需要可靠且可扩展的智能体基础设施。"
  },
  {
    "i": 26,
    "speaker": "Speaker 2",
    "text": "完全正确。"
  },
  {
    "i": 27,
    "speaker": "Speaker 2",
    "text": "而且这些任务越复杂，"
  },
  {
    "i": 28,
    "speaker": "Speaker 2",
    "text": "我们需要给予这些智能体更深入的访问权限"
  },
  {
    "i": 29,
    "speaker": "Speaker 2",
    "text": "为了让它们有效。"
  },
  {
    "i": 30,
    "speaker": "Speaker 2",
    "text": "有了像Fable这样的模型，我们真的需要这些来..."
  },
  {
    "i": 31,
    "speaker": "Speaker 2",
    "text": "他们能够完成非常复杂的任务。"
  },
  {
    "i": 32,
    "speaker": "Speaker 2",
    "text": "你无法运行一个有效的智能体"
  },
  {
    "i": 33,
    "speaker": "Speaker 2",
    "text": "在不授予其访问您凭证权限的情况下，"
  },
  {
    "i": 34,
    "speaker": "Speaker 2",
    "text": "你们的内部知识库或数据库。"
  },
  {
    "i": 35,
    "speaker": "Speaker 2",
    "text": "如果你想让这些智能体为你生成代码，"
  },
  {
    "i": 36,
    "speaker": "Speaker 2",
    "text": "你需要让他们访问你实际的代码库"
  },
  {
    "i": 37,
    "speaker": "Speaker 2",
    "text": "这样他们就可以推动PR并进入生产环境。"
  },
  {
    "i": 38,
    "speaker": "Speaker 2",
    "text": "最后，你需要授予它们身份和认证。"
  },
  {
    "i": 39,
    "speaker": "Speaker 2",
    "text": "我们的智能体越来越多地不仅仅作为Claude来行动，"
  },
  {
    "i": 40,
    "speaker": "Speaker 2",
    "text": "但对我来说或Jess，通过我们的电子邮件和Slack。"
  },
  {
    "i": 41,
    "speaker": "Speaker 2",
    "text": "当我们赋予智能体这些类似人类的能力时，"
  },
  {
    "i": 42,
    "speaker": "Speaker 2",
    "text": "我们期望它们会做出反应。"
  },
  {
    "i": 43,
    "speaker": "Speaker 2",
    "text": "我们期望它们以更加类似人类的方式与它们互动。"
  },
  {
    "i": 44,
    "speaker": "Speaker 2",
    "text": "互动的形式正在发生变化，"
  },
  {
    "i": 45,
    "speaker": "Speaker 2",
    "text": "而不仅仅是这些互动的持续时间。"
  },
  {
    "i": 46,
    "speaker": "Speaker 2",
    "text": "一些智能体非常善于对话。"
  },
  {
    "i": 47,
    "speaker": "Speaker 2",
    "text": "你可以引导它们，沿途提供指导。"
  },
  {
    "i": 48,
    "speaker": "Speaker 2",
    "text": "如果你认为它们偏离了轨道，甚至可以Interrupt它们。"
  },
  {
    "i": 49,
    "speaker": "Speaker 2",
    "text": "一些基于新模型（如Fable）的智能体"
  },
  {
    "i": 50,
    "speaker": "Speaker 2",
    "text": "非常注重结果。"
  },
  {
    "i": 51,
    "speaker": "Speaker 2",
    "text": "如果你有明确的信号或标准"
  },
  {
    "i": 52,
    "speaker": "Speaker 2",
    "text": "说明你希望看到什么被完成，"
  },
  {
    "i": 53,
    "speaker": "Speaker 2",
    "text": "你可以提供这些标准并让它进行迭代"
  },
  {
    "i": 54,
    "speaker": "Speaker 2",
    "text": "直到满足你的退出条件。"
  },
  {
    "i": 55,
    "speaker": "Speaker 2",
    "text": "最后，你可能几天前开始了一个任务"
  },
  {
    "i": 56,
    "speaker": "Speaker 2",
    "text": "然后想在很久之后再继续处理它。"
  },
  {
    "i": 57,
    "speaker": "Speaker 2",
    "text": "一个强大的智能体平台需要支持"
  },
  {
    "i": 58,
    "speaker": "Speaker 2",
    "text": "所有这些类型的交互模式。"
  },
  {
    "i": 59,
    "speaker": "Speaker 2",
    "text": "我们提供的基础设施和基本组件"
  },
  {
    "i": 60,
    "speaker": "Speaker 2",
    "text": "必须一次性把所有这些都提供给你"
  },
  {
    "i": 61,
    "speaker": "Speaker 2",
    "text": "同时还要非常非常灵活"
  },
  {
    "i": 62,
    "speaker": "Speaker 2",
    "text": "这样你就可以根据自己的需求进行定制。"
  },
  {
    "i": 63,
    "speaker": "Speaker 1",
    "text": "现在很明显，我们对智能体的期望很高。"
  },
  {
    "i": 64,
    "speaker": "Speaker 1",
    "text": "历史上，这意味着"
  },
  {
    "i": 65,
    "speaker": "Speaker 1",
    "text": "我们把负担都推给了你，开发者。"
  },
  {
    "i": 66,
    "speaker": "Speaker 1",
    "text": "在研究中，我们进行了"
  },
  {
    "i": 67,
    "speaker": "Speaker 1",
    "text": "在推出云托管智能体之前，"
  },
  {
    "i": 68,
    "speaker": "Speaker 1",
    "text": "我们发现开发者们确实渴望"
  },
  {
    "i": 69,
    "speaker": "Speaker 1",
    "text": "与我们一同攀登指数级增长的曲线，"
  },
  {
    "i": 70,
    "speaker": "Speaker 1",
    "text": "但他们在一些关键领域遇到了困难。"
  },
  {
    "i": 71,
    "speaker": "Speaker 1",
    "text": "首先，上下文管理。"
  },
  {
    "i": 72,
    "speaker": "Speaker 1",
    "text": "在正确的时间提供正确的上下文"
  },
  {
    "i": 73,
    "speaker": "Speaker 1",
    "text": "实际上非常难以调整"
  },
  {
    "i": 74,
    "speaker": "Speaker 1",
    "text": "尽管这完全必要。"
  },
  {
    "i": 75,
    "speaker": "Speaker 1",
    "text": "而在错误的时间提供上下文"
  },
  {
    "i": 76,
    "speaker": "Speaker 1",
    "text": "可能会严重分散智能体的注意力。"
  },
  {
    "i": 77,
    "speaker": "Speaker 1",
    "text": "我们有一半的开发者提到"
  },
  {
    "i": 78,
    "speaker": "Speaker 1",
    "text": "基础设施方面的担忧"
  },
  {
    "i": 79,
    "speaker": "Speaker 1",
    "text": "是他们最大的生产障碍。"
  },
  {
    "i": 80,
    "speaker": "Speaker 1",
    "text": "因此智能体会产生突发性的工作负载，"
  },
  {
    "i": 81,
    "speaker": "Speaker 1",
    "text": "它们具有不可预测的计算模式。"
  },
  {
    "i": 82,
    "speaker": "Speaker 1",
    "text": "在保证安全性的同时进行扩展非常困难"
  },
  {
    "i": 83,
    "speaker": "Speaker 1",
    "text": "同时还要满足延迟目标。"
  },
  {
    "i": 84,
    "speaker": "Speaker 1",
    "text": "最后，可观测性确实非常困难。"
  },
  {
    "i": 85,
    "speaker": "Speaker 1",
    "text": "你如何知道你的智能体"
  },
  {
    "i": 86,
    "speaker": "Speaker 1",
    "text": "是否产生了高质量的输出？"
  },
  {
    "i": 87,
    "speaker": "Speaker 1",
    "text": "这些是非确定性模型"
  },
  {
    "i": 88,
    "speaker": "Speaker 1",
    "text": "并且会产生大量非结构化数据。"
  },
  {
    "i": 89,
    "speaker": "Speaker 1",
    "text": "因此，云托管智能体应运而生。"
  },
  {
    "i": 90,
    "speaker": "Speaker 1",
    "text": "我们完成了平台工作，这样你就不用做了。"
  },
  {
    "i": 91,
    "speaker": "Speaker 1",
    "text": "托管智能体结合了基础设施、智能体基本组件，"
  },
  {
    "i": 92,
    "speaker": "Speaker 1",
    "text": "以及管理数据的能力。"
  },
  {
    "i": 93,
    "speaker": "Speaker 1",
    "text": "因此，云托管智能体应运而生。"
  },
  {
    "i": 94,
    "speaker": "Speaker 1",
    "text": "我们完成了平台工作，这样你就不用做了。"
  },
  {
    "i": 95,
    "speaker": "Speaker 1",
    "text": "托管智能体结合了基础设施、智能体基本组件，"
  },
  {
    "i": 96,
    "speaker": "Speaker 1",
    "text": "以及开箱即用的可观测性，"
  },
  {
    "i": 97,
    "speaker": "Speaker 1",
    "text": "所有这些都可以在云平台上的一个套餐中使用。"
  },
  {
    "i": 98,
    "speaker": "Speaker 1",
    "text": "在本次演示过程中，我们将详细探讨这些组件"
  },
  {
    "i": 99,
    "speaker": "Speaker 1",
    "text": "每一个部分。"
  },
  {
    "i": 100,
    "speaker": "Speaker 1",
    "text": "再加上几个演示。"
  },
  {
    "i": 101,
    "speaker": "Speaker 2",
    "text": "那么我们来谈谈最基本的基本构建模块"
  },
  {
    "i": 102,
    "speaker": "Speaker 2",
    "text": "云托管智能体。"
  },
  {
    "i": 103,
    "speaker": "Speaker 2",
    "text": "在最核心的部分，你有一个你定义的智能体。"
  },
  {
    "i": 104,
    "speaker": "Speaker 2",
    "text": "这将是系统提示，也就是你想要使用的模型，"
  },
  {
    "i": 105,
    "speaker": "Speaker 2",
    "text": "任何你想要加载的技能"
  },
  {
    "i": 106,
    "speaker": "Speaker 2",
    "text": "输入到你的智能体，以及具有权限的工具"
  },
  {
    "i": 107,
    "speaker": "Speaker 2",
    "text": "对于那些你希望该智能体具备的工具。"
  },
  {
    "i": 108,
    "speaker": "Speaker 2",
    "text": "这就像智能体的身份。"
  },
  {
    "i": 109,
    "speaker": "Speaker 2",
    "text": "接下来，你还有配置的环境。"
  },
  {
    "i": 110,
    "speaker": "Speaker 2",
    "text": "这就像一个模板，你可以在其中定义网络允许列表"
  },
  {
    "i": 111,
    "speaker": "Speaker 2",
    "text": "以及任何你可能想要预先安装的软件包。"
  },
  {
    "i": 112,
    "speaker": "Speaker 2",
    "text": "这就像智能体将要生活于其中的世界。"
  },
  {
    "i": 113,
    "speaker": "Speaker 2",
    "text": "你把那部分、环境和智能体，"
  },
  {
    "i": 114,
    "speaker": "Speaker 2",
    "text": "然后你用它来运行一个会话。"
  },
  {
    "i": 115,
    "speaker": "Speaker 2",
    "text": "会为您配置一个沙盒。"
  },
  {
    "i": 116,
    "speaker": "Speaker 2",
    "text": "我们设置好 harness，然后 Cloud 开始执行。"
  },
  {
    "i": 117,
    "speaker": "Speaker 2",
    "text": "任何您想要包含的凭证和资源"
  },
  {
    "i": 118,
    "speaker": "Speaker 2",
    "text": "在那个会话中，都是挂载到它上面的"
  },
  {
    "i": 119,
    "speaker": "Speaker 2",
    "text": "并且可供云平台使用。"
  },
  {
    "i": 120,
    "speaker": "Speaker 2",
    "text": "最后，我们有活动。"
  },
  {
    "i": 121,
    "speaker": "Speaker 2",
    "text": "这些是智能体生成的任何内容"
  },
  {
    "i": 122,
    "speaker": "Speaker 2",
    "text": "正如它在执行操作或任何你可能想要的事件"
  },
  {
    "i": 123,
    "speaker": "Speaker 2",
    "text": "为了引导智能体，需要提供上游信息。"
  },
  {
    "i": 124,
    "speaker": "Speaker 2",
    "text": "这就是你可能想要保持联系的方式。"
  },
  {
    "i": 125,
    "speaker": "Speaker 2",
    "text": "事件和状态感知正是我们能够发挥优势的地方"
  },
  {
    "i": 126,
    "speaker": "Speaker 2",
    "text": "提供一个平台，让你可以构建自己的产品"
  },
  {
    "i": 127,
    "speaker": "Speaker 2",
    "text": "在……之上，你可以使用诸如内存之类的原语"
  },
  {
    "i": 128,
    "speaker": "Speaker 2",
    "text": "以及其他我们拥有的功能"
  },
  {
    "i": 129,
    "speaker": "Speaker 2",
    "text": "真正优化这些智能体的性能。"
  },
  {
    "i": 130,
    "speaker": "Speaker 1",
    "text": "所以正如Michael提到的，活动是核心"
  },
  {
    "i": 131,
    "speaker": "Speaker 1",
    "text": "而智能体集成的核心所在。"
  },
  {
    "i": 132,
    "speaker": "Speaker 1",
    "text": "那么，让我们来实际解释一下这意味着什么。"
  },
  {
    "i": 133,
    "speaker": "Speaker 1",
    "text": "Managed Agents 中的每一件事都是基于事件的。"
  },
  {
    "i": 134,
    "speaker": "Speaker 1",
    "text": "因此这些是结构清晰的耐用转录文本"
  },
  {
    "i": 135,
    "speaker": "Speaker 1",
    "text": "帮助您跟踪智能体的进展。"
  },
  {
    "i": 136,
    "speaker": "Speaker 1",
    "text": "首先，有用户事件。"
  },
  {
    "i": 137,
    "speaker": "Speaker 1",
    "text": "这些实际上是发送给智能体以指导它的内容。"
  },
  {
    "i": 138,
    "speaker": "Speaker 1",
    "text": "接下来是智能体事件。"
  },
  {
    "i": 139,
    "speaker": "Speaker 1",
    "text": "这就是智能体实际在做的事情。"
  },
  {
    "i": 140,
    "speaker": "Speaker 1",
    "text": "所以这是消息传递。"
  },
  {
    "i": 141,
    "speaker": "Speaker 1",
    "text": "这是工具执行，上下文压缩，"
  },
  {
    "i": 142,
    "speaker": "Speaker 1",
    "text": "工程。"
  },
  {
    "i": 143,
    "speaker": "Speaker 1",
    "text": "甚至将任务委托给其他智能体。"
  },
  {
    "i": 144,
    "speaker": "Speaker 1",
    "text": "接下来是会话事件。"
  },
  {
    "i": 145,
    "speaker": "Speaker 1",
    "text": "这是你了解刚刚委托的任务"
  },
  {
    "i": 146,
    "speaker": "Speaker 1",
    "text": "进展的方式。"
  },
  {
    "i": 147,
    "speaker": "Speaker 1",
    "text": "所以这是整体生命周期，状态"
  },
  {
    "i": 148,
    "speaker": "Speaker 1",
    "text": "转换，错误和结果过程。"
  },
  {
    "i": 149,
    "speaker": "Speaker 1",
    "text": "最后是跨度事件。"
  },
  {
    "i": 150,
    "speaker": "Speaker 1",
    "text": "这个事件流中有很多内容。"
  },
  {
    "i": 151,
    "speaker": "Speaker 1",
    "text": "这有助于你将相关事件分组"
  },
  {
    "i": 152,
    "speaker": "Speaker 1",
    "text": "并以更聚合、更可测量的方式查看这些事件。"
  },
  {
    "i": 153,
    "speaker": "Speaker 2",
    "text": "那么让我们转向一个现实世界的例子，"
  },
  {
    "i": 154,
    "speaker": "Speaker 2",
    "text": "我们使用"
  },
  {
    "i": 155,
    "speaker": "Speaker 2",
    "text": "云托管智能体构建的某物，即Pascal，它"
  },
  {
    "i": 156,
    "speaker": "Speaker 2",
    "text": "使用一个假设的在线杂货配送服务订单"
  },
  {
    "i": 157,
    "speaker": "Speaker 2",
    "text": "数据来分析这些数据"
  },
  {
    "i": 158,
    "speaker": "Speaker 2",
    "text": "并为我们的团队提供见解。"
  },
  {
    "i": 159,
    "speaker": "Speaker 2",
    "text": "智能体可以在几分钟内生成分析报告，"
  },
  {
    "i": 160,
    "speaker": "Speaker 2",
    "text": "利用一个预加载的数据集和一组Python软件包"
  },
  {
    "i": 161,
    "speaker": "Speaker 2",
    "text": "以及我们安装并上传到容器中的脚本"
  },
  {
    "i": 162,
    "speaker": "Speaker 2",
    "text": "从而让它能够运行。"
  },
  {
    "i": 163,
    "speaker": "Speaker 2",
    "text": "你可以在Cloud控制台中看到每一个事件，甚至可以聊天"
  },
  {
    "i": 164,
    "speaker": "Speaker 2",
    "text": "与调试智能体，以进一步优化"
  },
  {
    "i": 165,
    "speaker": "Speaker 2",
    "text": "你的集成。"
  },
  {
    "i": 166,
    "speaker": "Speaker 2",
    "text": "那么进入视频本身，这个"
  },
  {
    "i": 167,
    "speaker": "Speaker 2",
    "text": "就是Pascal的主页。"
  },
  {
    "i": 168,
    "speaker": "Speaker 2",
    "text": "我们可以启动一个智能体会话，Cloud就会开始运行。"
  },
  {
    "i": 169,
    "speaker": "Speaker 2",
    "text": "它需要几分钟的时间才能完成。"
  },
  {
    "i": 170,
    "speaker": "Speaker 2",
    "text": "所以在它运行的同时，我们会转到开发者控制台"
  },
  {
    "i": 171,
    "speaker": "Speaker 2",
    "text": "并且实际上可以看到Cloud"
  },
  {
    "i": 172,
    "speaker": "Speaker 2",
    "text": "实时生成的事件，并能够更好地理解"
  },
  {
    "i": 173,
    "speaker": "Speaker 2",
    "text": "Cloud正在实时执行的操作。"
  },
  {
    "i": 174,
    "speaker": "Speaker 2",
    "text": "我们还可以查看智能体配置"
  },
  {
    "i": 175,
    "speaker": "Speaker 2",
    "text": "以及为这次会话设置的环境配置。"
  },
  {
    "i": 176,
    "speaker": "Speaker 2",
    "text": "所以这个实例，这就是系统提示"
  },
  {
    "i": 177,
    "speaker": "Speaker 2",
    "text": "以及这就是模型。"
  },
  {
    "i": 178,
    "speaker": "Speaker 2",
    "text": "然后转到环境配置部分，"
  },
  {
    "i": 179,
    "speaker": "Speaker 2",
    "text": "你可以看到允许的网络权限"
  },
  {
    "i": 180,
    "speaker": "Speaker 2",
    "text": "以及这些包。"
  },
  {
    "i": 181,
    "speaker": "Speaker 2",
    "text": "回到实际的页面本身，"
  },
  {
    "i": 182,
    "speaker": "Speaker 2",
    "text": "看起来我们的分析基本上已经完成"
  },
  {
    "i": 183,
    "speaker": "Speaker 2",
    "text": "现在我们可以深入探讨Cloud发现的见解。"
  },
  {
    "i": 184,
    "speaker": "Speaker 2",
    "text": "从产品角度来看，似乎香蕉真的很"
  },
  {
    "i": 185,
    "speaker": "Speaker 2",
    "text": "受欢迎，还有其他一些产品"
  },
  {
    "i": 186,
    "speaker": "Speaker 2",
    "text": "也同样很受欢迎，但每个人都应该获得"
  },
  {
    "i": 187,
    "speaker": "Speaker 2",
    "text": "他们的香蕉。"
  },
  {
    "i": 188,
    "speaker": "Speaker 2",
    "text": "从客户洞察来看，我们发现周日下午"
  },
  {
    "i": 189,
    "speaker": "Speaker 2",
    "text": "在线订单非常受欢迎。"
  },
  {
    "i": 190,
    "speaker": "Speaker 2",
    "text": "所以你可能想要将订单安排在"
  },
  {
    "i": 191,
    "speaker": "Speaker 2",
    "text": "一天中的其他时间。"
  },
  {
    "i": 192,
    "speaker": "Speaker 2",
    "text": "最后，我们设置了一个预测"
  },
  {
    "i": 193,
    "speaker": "Speaker 2",
    "text": "模拟器，让我们分析客户是否更有可能"
  },
  {
    "i": 194,
    "speaker": "Speaker 2",
    "text": "再次订购商品。"
  },
  {
    "i": 195,
    "speaker": "Speaker 2",
    "text": "回到开发者控制台，"
  },
  {
    "i": 196,
    "speaker": "Speaker 2",
    "text": "我们可以在侧边栏看到"
  },
  {
    "i": 197,
    "speaker": "Speaker 2",
    "text": "我们可以启动对会话本身的分析，其中"
  },
  {
    "i": 198,
    "speaker": "Speaker 2",
    "text": "Cloud会查看所有事件"
  },
  {
    "i": 199,
    "speaker": "Speaker 2",
    "text": "并进一步为我们提供见解"
  },
  {
    "i": 200,
    "speaker": "Speaker 2",
    "text": "关于如何优化我们构建的集成。"
  },
  {
    "i": 201,
    "speaker": "Speaker 2",
    "text": "在这一特定情况下，我们看到"
  },
  {
    "i": 202,
    "speaker": "Speaker 2",
    "text": "我们提供给 Cloud 的一些脚本实际上"
  },
  {
    "i": 203,
    "speaker": "Speaker 2",
    "text": "真的，真的很慢。"
  },
  {
    "i": 204,
    "speaker": "Speaker 2",
    "text": "因此，我们可能需要优化我们编写的 Python 代码"
  },
  {
    "i": 205,
    "speaker": "Speaker 2",
    "text": "为了让他们做得更好。"
  },
  {
    "i": 206,
    "speaker": "Speaker 2",
    "text": "所以我们刚刚看了几件事。"
  },
  {
    "i": 207,
    "speaker": "Speaker 2",
    "text": "我们查看了开发者控制台，"
  },
  {
    "i": 208,
    "speaker": "Speaker 2",
    "text": "你看到了实际效果。"
  },
  {
    "i": 209,
    "speaker": "Speaker 2",
    "text": "但事实上有很多不同的方法"
  },
  {
    "i": 210,
    "speaker": "Speaker 2",
    "text": "帮助您今天开始使用云托管智能体"
  },
  {
    "i": 211,
    "speaker": "Speaker 2",
    "text": "为了与你创建的这些智能体进行交互。"
  },
  {
    "i": 212,
    "speaker": "Speaker 2",
    "text": "你只需选择最适合你工作流程的那个。"
  },
  {
    "i": 213,
    "speaker": "Speaker 2",
    "text": "我最喜欢的是云API技能"
  },
  {
    "i": 214,
    "speaker": "Speaker 2",
    "text": "我们今天在Cloud Code中可用的内容。"
  },
  {
    "i": 215,
    "speaker": "Speaker 2",
    "text": "你只需要问问Cloud，嘿，"
  },
  {
    "i": 216,
    "speaker": "Speaker 2",
    "text": "我想开始使用云托管智能体。"
  },
  {
    "i": 217,
    "speaker": "Speaker 2",
    "text": "并且它将帮助你完成入职流程"
  },
  {
    "i": 218,
    "speaker": "Speaker 2",
    "text": "集成到您现有的代码库中。"
  },
  {
    "i": 219,
    "speaker": "Speaker 2",
    "text": "接下来，我们还有ANSI盟友，"
  },
  {
    "i": 220,
    "speaker": "Speaker 2",
    "text": "这是我们最近发布的。"
  },
  {
    "i": 221,
    "speaker": "Speaker 2",
    "text": "这使得交互变得非常容易"
  },
  {
    "i": 222,
    "speaker": "Speaker 2",
    "text": "通过我们的API进行脚本编写和CI，"
  },
  {
    "i": 223,
    "speaker": "Speaker 2",
    "text": "CDL。"
  },
  {
    "i": 224,
    "speaker": "Speaker 2",
    "text": "最后，我们还有我们的开发者码头"
  },
  {
    "i": 225,
    "speaker": "Speaker 2",
    "text": "以及我们的食谱。"
  },
  {
    "i": 226,
    "speaker": "Speaker 2",
    "text": "这些提供了很多实际的例子"
  },
  {
    "i": 227,
    "speaker": "Speaker 2",
    "text": "并且提供可以直接复制粘贴的最常见模式示例"
  },
  {
    "i": 228,
    "speaker": "Speaker 2",
    "text": "这是我们为云托管智能体准备的。"
  },
  {
    "i": 229,
    "speaker": "Speaker 1",
    "text": "那么我们现在已经介绍了基础知识，"
  },
  {
    "i": 230,
    "speaker": "Speaker 1",
    "text": "我想谈一谈一些更高级的功能"
  },
  {
    "i": 231,
    "speaker": "Speaker 1",
    "text": "我们最近已经上线了。"
  },
  {
    "i": 232,
    "speaker": "Speaker 1",
    "text": "那么首先，多智能体协调。"
  },
  {
    "i": 233,
    "speaker": "Speaker 1",
    "text": "在这里，Cloud能够委派任务"
  },
  {
    "i": 234,
    "speaker": "Speaker 1",
    "text": "与其他具有独立上下文窗口的智能体进行交互，"
  },
  {
    "i": 235,
    "speaker": "Speaker 1",
    "text": "使其能够并行化"
  },
  {
    "i": 236,
    "speaker": "Speaker 1",
    "text": "日益复杂的工作。"
  },
  {
    "i": 237,
    "speaker": "Speaker 1",
    "text": "在结果方面，Cloud 会根据预定义的退出标准进行迭代"
  },
  {
    "i": 238,
    "speaker": "Speaker 1",
    "text": "或者直到它满足其目标为止的评分标准。"
  },
  {
    "i": 239,
    "speaker": "Speaker 1",
    "text": "你负责这个目标，以及Cloud"
  },
  {
    "i": 240,
    "speaker": "Speaker 1",
    "text": "负责完成它。"
  },
  {
    "i": 241,
    "speaker": "Speaker 1",
    "text": "有了记忆功能，Cloud 就能够读写记忆存储。"
  },
  {
    "i": 242,
    "speaker": "Speaker 1",
    "text": "默认情况下，没有记忆功能，Cloud"
  },
  {
    "i": 243,
    "speaker": "Speaker 1",
    "text": "每次会议都会从头开始。"
  },
  {
    "i": 244,
    "speaker": "Speaker 1",
    "text": "但是有了记忆，它就具备了对之前运行情况的意识"
  },
  {
    "i": 245,
    "speaker": "Speaker 1",
    "text": "并且下次可以做得更好。"
  },
  {
    "i": 246,
    "speaker": "Speaker 1",
    "text": "梦境是建立在记忆之上的，"
  },
  {
    "i": 247,
    "speaker": "Speaker 1",
    "text": "在这里，Claude 正在进行反思并将其系统化"
  },
  {
    "i": 248,
    "speaker": "Speaker 1",
    "text": "关于学习和将其编码为新的记忆。"
  },
  {
    "i": 249,
    "speaker": "Speaker 1",
    "text": "并且这确保它可以持续启动"
  },
  {
    "i": 250,
    "speaker": "Speaker 1",
    "text": "具有更精炼、更优化和更精选的记忆集。"
  },
  {
    "i": 251,
    "speaker": "Speaker 1",
    "text": "所有这些都非常令人兴奋的智能体功能，"
  },
  {
    "i": 252,
    "speaker": "Speaker 1",
    "text": "我们非常兴奋能够提高上限"
  },
  {
    "i": 253,
    "speaker": "Speaker 1",
    "text": "关于智能体能够产生什么。"
  },
  {
    "i": 254,
    "speaker": "Speaker 1",
    "text": "然而，我们也听到需要与您见面"
  },
  {
    "i": 255,
    "speaker": "Speaker 1",
    "text": "你所在的位置，这意味着要进行"
  },
  {
    "i": 256,
    "speaker": "Speaker 1",
    "text": "使我们的基础设施更加模块化。"
  },
  {
    "i": 257,
    "speaker": "Speaker 1",
    "text": "所以使用自托管沙盒，你可以运行智能体循环"
  },
  {
    "i": 258,
    "speaker": "Speaker 1",
    "text": "并且可以在你的基础设施中直接执行工具"
  },
  {
    "i": 259,
    "speaker": "Speaker 1",
    "text": "因此文件和软件包永远不会离开你的边界。"
  },
  {
    "i": 260,
    "speaker": "Speaker 1",
    "text": "通过MCP隧道，Claude可以访问私有的MCP服务器"
  },
  {
    "i": 261,
    "speaker": "Speaker 1",
    "text": "你不希望暴露在开放互联网上的内容。"
  },
  {
    "i": 262,
    "speaker": "Speaker 1",
    "text": "所有这些东西的构建目的都是为了让你能够部署"
  },
  {
    "i": 263,
    "speaker": "Speaker 1",
    "text": "在你们的企业内部，按照你们自身的安全原则。"
  },
  {
    "i": 264,
    "speaker": "Speaker 2",
    "text": "除了所有这些令人惊叹的功能之外"
  },
  {
    "i": 265,
    "speaker": "Speaker 2",
    "text": "我们已经拥有的，就像凯特林今天早些时候提到的那样，"
  },
  {
    "i": 266,
    "speaker": "Speaker 2",
    "text": "我们刚刚发布了两个全新且非常令人兴奋的功能。"
  },
  {
    "i": 267,
    "speaker": "Speaker 2",
    "text": "一个是计划部署，它允许你设置"
  },
  {
    "i": 268,
    "speaker": "Speaker 2",
    "text": "我们触发新会话的重复计划"
  },
  {
    "i": 269,
    "speaker": "Speaker 2",
    "text": "代表您处理您可能有的任何重复性工作。"
  },
  {
    "i": 270,
    "speaker": "Speaker 2",
    "text": "然后接下来我们在Vaults中还有环境变量，"
  },
  {
    "i": 271,
    "speaker": "Speaker 2",
    "text": "允许您提供安全的凭证"
  },
  {
    "i": 272,
    "speaker": "Speaker 2",
    "text": "对于任何您可能希望 Claude 调用的 API 或 CLI"
  },
  {
    "i": 273,
    "speaker": "Speaker 2",
    "text": "在没有实际承担任何风险的情况下"
  },
  {
    "i": 274,
    "speaker": "Speaker 2",
    "text": "关于Claude看到实际的秘密词元。"
  },
  {
    "i": 275,
    "speaker": "Speaker 2",
    "text": "稍微深入探讨一下环境变量是如何工作的"
  },
  {
    "i": 276,
    "speaker": "Speaker 2",
    "text": "在 Vaults 中，我们放置一个不透明的占位符词元"
  },
  {
    "i": 277,
    "speaker": "Speaker 2",
    "text": "Claude 可以访问的容器本身内部。"
  },
  {
    "i": 278,
    "speaker": "Speaker 2",
    "text": "每当 Claude 尝试调用 API 或使用 CLI 时，"
  },
  {
    "i": 279,
    "speaker": "Speaker 2",
    "text": "它将只是以这种方式使用该环境变量"
  },
  {
    "i": 280,
    "speaker": "Speaker 2",
    "text": "它将使用任何其他环境变量。"
  },
  {
    "i": 281,
    "speaker": "Speaker 2",
    "text": "当该网络请求被发出时，"
  },
  {
    "i": 282,
    "speaker": "Speaker 2",
    "text": "我们将在此请求发出时注入真实的秘密值"
  },
  {
    "i": 283,
    "speaker": "Speaker 2",
    "text": "这样Claude就永远不会实际看到这些值。"
  },
  {
    "i": 284,
    "speaker": "Speaker 2",
    "text": "这就是秘密令牌的真正价值。"
  }
]