内容来源：https://aibusiness.com/agentic-ai/anthropic-rolls-out-claude-duet-enterprises-aren-t-ready

内容总结：

Anthropic本周发布研究预览版个人AI代理应用Claude Duet，正式加入个人AI代理市场竞争。该应用基于Claude Opus 5.5构建，具备主动执行任务能力，用户离开后仍可继续完成工作，但代理在代表用户操作前会先进行确认。Anthropic表示，Duet面向专业人士和小企业主，提供iOS和Android独立应用、Claude.ai网页版，并纳入Pro和Max订阅计划。

Duet的发布使Anthropic成为最新一家进入个人代理市场的AI厂商。OpenAI于9月推出个人代理Dots，谷歌于10月8日发布Gemini Agent。三家厂商的发布均紧随Meta旗下个人AI代理Muse的迅速走红，后者可管理邮件发送、旅行预订、购物及日常日程等任务，Meta还试图将Muse拓展至小企业应用场景。

Forrester Research分析师William McKeon-White表示，Duet的发布虽略显迟缓，但Anthropic采取了比竞争对手更为谨慎的策略。他指出，Duet并无特别独特之处，区别在于其引入方式上的克制。Anthropic称，Duet内置专用分类器和安全系统提示，可防止敏感信息泄露，拒绝共享信用卡号和政府身份证件等请求。用户可通过Stripe数字钱包指示Duet为企业购物，但Duet被限制执行证券交易或账户间转账。McKeon-White认为，对于这类新型代理，谨慎 approach 更为可取，业界仍在适应具有如此自主程度的代理，互联网上不乏意外和失控行为的案例。

然而，企业级市场尚未准备好迎接个人AI代理。GAI Insights首席执行官兼分析师Paul Baier表示，B2B客户不会在数月内推出Duet，可能需要数年，甚至永远不会推出。他指出，许多企业甚至尚未启用Anthropic于1月发布的以非主动方式为主的代理Claude Cowork。Baier称，目前没有大型企业客户对Duet表示兴趣或计划在未来六个月内使用。

企业面临的挑战在于，允许非技术员工使用AI聊天和项目功能风险较低，但使用Cowork和Duet则更为复杂，因为个人代理可自动化处理机密信息并连接IT系统。Baier表示，每个IT、治理和安全部门都在强调，在允许上千名员工使用机密数据并接入现有IT系统进行“氛围编程”或快速原型开发之前，需要更加谨慎。他补充称，没有任何员工规模超过1000人的公司急于在第四季度部署Dots、Muse或Duet。

中文翻译：

由Google Cloud赞助
选择你的首批生成式AI用例
要开始使用生成式AI，首先应聚焦于能够改善人类信息体验的领域。
这家生成式AI供应商发布了该应用，并配备了严格的安全防护措施，以适用于专业和小型企业场景。
Anthropic本周加入了AI个人智能体热潮，推出了研究预览版的Claude Duet。
然而，尽管所有主要前沿模型制造商现在都提供个人工作智能体，企业距离使用这些智能体或让员工访问它们还差得很远。
Duet是一款构建在Claude Opus 5.5之上的主动式智能体应用。通过该应用，即使用户离开，Claude也能完成任务；不过，智能体在代表用户执行操作之前会先进行检查。
Anthropic表示，Duet是为专业人士和小企业主打造的。它以独立应用的形式在iOS和Android上提供，也可通过Claude.ai在网页端使用，并且包含在Pro和Max订阅计划中。
借助Duet，Anthropic成为最新一家进入个人智能体市场的AI供应商。OpenAI于9月推出了其个人智能体Dots，Google于10月8日发布了Gemini Agent。
这三款智能体的发布都紧随Meta Muse的一炮而红。Meta Muse是这家社交媒体巨头推出的个人AI智能体，可管理发送邮件、预订旅行、购物和处理日常日程等任务。Meta还试图将Muse引向小企业应用场景。
Forrester Research分析师William McKeon-White表示，虽然Duet的发布似乎稍显滞后，但该供应商采取了比竞争对手更为谨慎的方式。
“我不会说Duet有什么特别独特之处，除了它推出时所表现出的克制，”他说。
Anthropic指出，Duet包含专用的分类器和安全系统提示词，可防止敏感信息泄露。分类器和提示词会拒绝共享敏感信息的请求，例如信用卡号和政府身份证件。用户可以通过Stripe的数字钱包指示Duet为其企业购买物品。Duet被限制不允许用户要求其交易证券或在账户之间转移资金。
“坦白说，对于这些新型智能体来说，这是一种更好的方式，”McKeon-White说。“我们还在适应具有这种自主程度的智能体，而互联网上充斥着意外和失控行为的例子。”
此外，GAI Insights首席执行官兼分析师Paul Baier表示，企业尚未为个人AI智能体做好准备。
“B2B客户不会在数月内、也许数年内、也许永远不会推出Duet，”Baier说。他还补充说，许多企业甚至还没有启用Claude Cowork——Anthropic于1月发布的以非主动式为主的智能体。
“我们没有任何大型企业客户对使用它感到兴奋，或计划在未来六个月内使用它，”Baier在谈到Duet时说。
企业面临的部分挑战在于，虽然让非技术员工访问AI聊天和项目风险较低，但使用Cowork和Duet则变得更加复杂，因为个人智能体现在可以自动化处理机密信息并连接IT系统的流程。
“每个IT和治理团队以及安全团队都在说，‘嘿，等一下。在我们让1000名员工开始氛围编程或氛围原型开发使用机密数据并能接入我们现有IT系统的应用之前，我们需要更加谨慎，’”Baier补充道。“没有哪家超过1000人的公司在第四季度急于推出Dots、Muse或Duet。”

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The generative AI vendor released the app with stringent security guardrails for using it in professional and small business settings.
Anthropic joined the AI personal agent craze this week with the release of Claude Duet in research preview.
However, while all major frontier model makers now offer a personal work agent, enterprises are still far from using these agents or giving their employees access to them.
Duet is a proactive agent app built on Claude Opus 5.5. With the app, Claude completes tasks even when a user walks away; however, the agent checks before working on the user’s behalf.
Anthropic said it built Duet for professionals and small business owners. It is available as a standalone app for iOS and Android, on the web through Claude.ai, and as part of the Pro and Max plans.
With Duet, Anthropic becomes the latest AI vendor to enter the personal agent market. OpenAI introduced its personal agent, Dots, in September, and Google released Gemini Agent on Oct. 8.
All three agent releases followed the instant success of Meta Muse, a personal AI agent from the social media giant that manages tasks such as sending emails, booking travel, shopping and handling daily schedules. Meta is also trying to point Muse toward small business applications.
While the Duet release appears a bit late, the vendor is taking a more cautious approach than competitors, said William McKeon-White, an analyst at Forrester Research
“I wouldn't say there's anything super unique about Duet, except the restraint in which it's being introduced,” he said.
Anthropic noted that Duet includes a dedicated classifier and safety-system prompts that guard against leakage of sensitive information. The classifier and prompt reject requests to share sensitive information, such as credit card numbers and government IDs. Users can direct Duet to buy things for their businesses using Stripe's digital wallet. Duet is restricted from users asking it to trade securities or move money among accounts.
“With these new types of agents, frankly that's a better approach,” McKeon-White said. “We're still getting used to agents with this degree of autonomy, and the internet is littered with examples of unintended and rogue behavior.”
Moreover, enterprises aren't yet ready for personal AI agents, said Paul Baier, CEO and analyst at GAI Insights.
“B2B customers aren’t going to be rolling out Duet for months, maybe years, maybe never rolling it out,” Baier said. He added that many enterprises have not even switched on Claude Cowork, the mainly non-proactive agent Anthropic released in January.
“We have no large corporate customers who are excited to use it or plan to use it in the next six months,” Baier said, referring to Duet.
Part of the challenge for enterprises is that while allowing their non-technical employees access to AI chats and projects is low risk, using Cowork and Duet becomes more complicated because personal agents can now automate processes that handle confidential information and connect to IT systems.
“Every IT and governance group and security group is saying, "Hey, wait a minute. We need to be more cautious before we let 1000 employees start vibe coding or vibe prototyping apps that use confidential data and can tap into our existing IT systems,’” Baier added. “No company over 1,000 people is clamoring to roll out Dots, Muse or Duet in Q4.”