Anthropic 正为企业客户提供一种新方式，以审查其员工如何使用该公司的人工智能聊天机器人 Claude —— 以及 AI 工作给他们带来了多少成本。这家 AI 巨头上周宣布为 Claude Enterprise 用户推出“智能报告”。这一交互式图表工具旨在分析团队 AI 会话记录的样本，并将工作归类，类别包括员工使用 Claude 完成的任务类型、其生成的输出内容以及成本。
Anthropic is giving corporate customers a new way to scrutinize how their employees use its artificial intelligence chatbot Claude — and how much the AI work is costing them. The AI giant announced "smart reports" for Claude Enterprise users last week. The interactive chart tool is designed to analyze a sample of a team's AI session transcripts and group work into categories, including the types of tasks employees use Claude for, the outputs it generates, and the cost.
根据 Claude 支持网页介绍，这些报告还能 —— 这无疑是让团队坐直了身子的部分 —— 向雇主标记“低效环节”，包括产生“无工作价值”或属于“个人或离题”的会话。实际上，智能报告工具为配备了 Claude Enterprise 的公司提供了一份某种意义上的 AI 成绩单。每个组织每月最多可免费运行 10 份报告。Anthropic 在其支持文档中强调了一个“重要”提示：“智能报告可帮助您了解采用情况并规划在 Claude 上的投资。它们并非为此设计，也不应用于评估个人绩效或做出雇佣决策。”
The reports can also — and this is the part that is undoubtedly making teams sit up straight — flag "inefficiencies" to employers, including sessions that produced "no work value" or were "personal or off-topic," according to a Claude support webpage. In effect, the smart reports tool offers companies equipped with Claude Enterprise something of an AI report card. Each organization can run up to 10 reports a month at no charge. One caveat that Anthropic highlighted as "important" in its support documentation: "Smart reports help you understand adoption and plan your investment in Claude. They aren't designed and should not be used for evaluating individual performance or making employment decisions."
要生成报告，管理者可选择团队、最长达过去 28 天的时间范围、要包含的 Claude 产品、分析重点领域，以及他们希望 Claude 在分析记录时回答的任何自定义问题。“Claude 随后会阅读该范围内的记录样本，将其归类为工作流和输出类型，为每组关联支出，并撰写发现摘要，”Claude 支持页面称。据 Anthropic 介绍，涉及“受限主题”的会话 —— 包括与种族、族裔、健康、法律问题、工会活动、政治或宗教观点相关的个人对话 —— “仅用于汇总分析，不包含摘要”。“个人对话也不包含个人会话或姓名，”Claude 支持页面称。
To produce a report, bosses can select a team, a time range of up to the last 28 days, the Claude products to include, focus areas for analysis, and any custom questions they want Claude to answer while analyzing the transcripts. "Claude then reads a sample of transcripts in that scope, groups them into workstreams and types of outputs, attaches spend to each group, and writes up what it found," the Claude support page says. According to Anthropic, sessions involving "restricted topics" — including personal conversations related to race, ethnicity, health, legal issues, union activity, and political or religious views — are "used only for aggregate analysis and don't include summaries." "Personal conversations also don't include individual sessions or names," the Claude support page says.
智能报告的推出正值许多公司试图控制AI成本并衡量任何生产力提升之际。一些公司正在纠结该在多大范围内允许员工使用AI工具。《彭博社》周一报道，高盛在评估了生产力和风险顾虑后，最近将Anthropic的Claude Code和微软GitHub编码工具的使用权限仅限工程师。据报道，这一举措激怒了一些使用这些工具工作的高盛交易员。
The launch of smart reports comes as many companies are trying to get a handle on AI costs and measure any productivity gains. Some are wrestling with how broadly to let employees use AI tools. Bloomberg reported Monday that Goldman Sachs recently limited access to Anthropic's Claude Code and Microsoft's GitHub coding tools to engineers only after assessing productivity and risk concerns. The move reportedly rankled some Goldman traders who used the tools for their work.