内容来源：https://aibusiness.com/agentic-ai/google-s-workplace-agent-is-out

内容总结：

谷歌近日在其于山景城举办的“Gemini at Work 2026”大会上正式推出旗下首款职场个人智能体——Gemini Agent。该产品可接入企业业务场景，辅助完成知识工作，包括回答问题、生成内容和编写代码，用户仅需通过一个提示框即可操作。与此同时，谷歌还发布了一项新的工作流集成功能，可直接在Google Workspace内部运行，并连接Docs、Drive、Sheets等广泛使用的企业生产力应用。谷歌由此成为首家推出可连接企业工作流智能体的前沿AI厂商，在AI竞赛中形成差异化定位。

Omdia分析师Mark Beccue将这一方向形容为“数字办公空间的争夺”，认为这是个人助理在日常工作中的实际落地。他指出，尽管谷歌是首家推出职场个人智能体的前沿AI开发商，但Meta和微软也已推出各自的数字员工产品——Meta上月发布了面向消费者的Muse个人智能体，可执行长期任务；微软则推出了面向企业团队的AI数字队友Copilot Autopilot。

然而，这一趋势也引发企业对数据安全的担忧。智能体需要访问企业数据才能发挥作用，而近期围绕AI智能体失控的焦虑可能使部分企业持谨慎态度。Futurum Group分析师David Nicholson认为，企业必须放下部分疑虑才能释放AI的真正价值。他表示，已在使Google Workspace的企业大概率不会介意Gemini Agent访问员工的Gmail、Google Slides和日历应用，但企业还需考虑存储在SAP、Salesforce、Oracle等环境中的数据。此外，即便企业允许访问，权限问题也随之而来——员工在微软平台上有时未经特定审批便无法访问谷歌文件。Beccue指出，这并非谷歌文档之间的互操作性问题，而是企业层面的权限管理问题，“安全正在侵蚀生产力”已成为普遍现象，智能体如何应对这一挑战尚待观察。

成本是另一大挑战。谷歌表示企业可在Gemini Agent中选择使用Gemini模型或Anthropic Claude，但部分模型运行成本高昂，且让用户自行选择模型需要高水平的管理能力和对token成本财务的深入了解。Beccue质疑，这是否意味着每位员工都要承担财务管理责任，企业又将如何应对。

Tekonyx创始人兼分析师Sid Nag指出，此类挑战表明，在复杂的多应用环境中部署工作流智能体，首次尝试很难成功，要达到精确程度尚需时日。Nicholson则表示，已处于谷歌环境中的企业可能选择Gemini Agent，而其他企业可能更倾向于将数据掌控在自己手中，以避免被大厂锁定。

尽管存在所有权和权限方面的诸多疑问，Nag认为谷歌此举方向正确，是将前沿能力与可执行的智能体能力相结合的优秀举措。谷歌表示，Gemini Agent可从网页、iOS和Android等任意设备访问，在云端运行，提供统一的记忆、上下文和个性化图谱，并可创建子智能体来执行特定任务。

中文翻译：

由谷歌云赞助
选择你的首批生成式AI用例
要开始使用生成式AI，首先应关注那些能够改善人类信息体验的领域。
该代理可以利用企业业务上下文来从事知识工作。然而，关于成本以及与其他工具集成的问题，对企业来说可能并不容易解决。
谷歌今天推出了Gemini Agent，这是其首个工作场所个人代理。但企业——即便是那些已经身处这家科技巨头生态系统中的企业——也需要自问，它们是否已准备好让AI代理访问其数据和上下文，以及这些代理能否处理员工早已习惯管理的日常流程和权限。
Gemini Agent被配置了企业的业务上下文，并可将这些信息用于知识工作，回答问题、创建内容以及编写代码——所有这些都可从一个单一的提示框中完成。
谷歌在位于山景城的“Gemini at Work 2026”大会上发布了该代理，同时还推出了一项新的工作流集成，可直接在Google Workspace内运行，并连接到其广泛使用的企业生产力应用，如Docs、Drive和Sheets。
该厂商是首家发布可连接企业工作流的代理的前沿AI提供商，这表明这家云服务提供商正如何在AI竞赛中实现差异化。
“这是一场围绕数字工作空间的布局，”Informa TechTarget旗下Omdia的分析师Mark Beccue表示。“这是个人助理在日常工作中的日常应用。”
虽然谷歌是首家推出工作场所个人代理的前沿AI开发者，但Meta和微软等竞争对手也已推出各自的数字员工。Meta上月推出了Muse个人代理；它为消费者执行长时间运行的任务，并可能对小型企业具有吸引力。与此同时，微软推出了Copilot Autopilot，这是面向企业团队的AI数字队友。
“这是进军数字工作空间最容易摘到的果实，”Beccue说。“这实际上是AI界面的现实化。我们将告诉AI我们想做什么，而它会去做。这就是我们的前进方向。”
然而，朝这个方向发展，谷歌、微软，甚至Meta，正迈向一个AI代理需要访问企业数据的世界。鉴于近期人们对AI代理失控的担忧，一些企业可能会对允许Gemini或其他模型访问这些数据持怀疑态度。
不过，Futurum Group的分析师David Nicholson表示，企业必须放下部分怀疑，才能认识到AI的力量。
“如果企业不打开保险库，让代理进入其中、四处挖掘并获得访问权限，我们就无法释放AI对企业真正价值的潜能，”他说。
他补充说，虽然一些企业可能会通过只允许Gemini Agent访问其组织的某些方面来进行试点，但另一些企业可能会全面投入。而且，已经在使用Google Workspace的企业，很可能不会介意允许Gemini Agent访问员工的Gmail、Google Slides和Calendar应用。
“所有这些都提供了基础性上下文，而这似乎是你为某些类型的运营所必须开放的最低门槛，”Nicholson说。
“不过，下一个问题是，‘那我的那些与谷歌无关的东西怎么办？’”他补充道，并指出企业还需要考虑存放在SAP、Salesforce、Oracle或其他大型企业环境中的数据。
此外，即便企业允许Gemini Agent访问这些其他环境，也会出现权限和访问的问题。例如，如果该代理试图访问SAP或微软，它是否有权限继续？使用微软平台的企业员工有时在没有特别批准的情况下会被限制访问谷歌文件。
“这并不是Google Docs之间的互操作性问题；这是公司设定的权限问题，”Beccue说。“这是一个普遍主题，你可以说安全性正在让我们付出生产力代价。代理将如何应对这一点？因为这些是组织设定的权限，而不是供应商设定的权限。”
Beccue说，另一个挑战是成本。谷歌表示，企业在使用Gemini Agent时可以在Gemini模型和Anthropic Claude之间选择。然而，有些模型的运行成本很高。此外，Beccue说，让用户选择他们使用的模型，需要高水平的管理以及对令牌成本财务的了解。
“这意味着我们正在学习如何成为代理提示者或助理提示者。现在……每个人都得负责财务了吗？”他继续说道。“组织将如何处理这一切？”
Tekonyx创始人兼分析师Sid Nag表示，诸如此类的挑战暴露出，“当你开始部署工作流代理时，你不可能第一次就成功，尤其是在复杂的多应用环境中，这种代理将要在其中弥合差距。”“要达到精确无误的水平，需要一些时间。”
Nicholson说，虽然已经处于谷歌环境中的组织可能会选择Gemini Agent，但其他组织可能会选择让数据更贴近自身。
“这个想法是，也许你希望把所有这些知识产权作为一个独立的东西来拥有，使其与某个可能将你锁定的巨头脱钩，”他说。
Nag说，尽管围绕所有权和权限存在所有这些疑问，谷歌的发布仍是朝着正确方向迈出的一步。
“这是一个很棒的 offering，也是谷歌很棒的一步，将其前沿能力与可执行的代理能力整合在一起，”他说。“这是一个覆盖层中间件API，”他说。
谷歌表示，Gemini Agent可从任何设备访问，包括网页、iOS和基于Android的移动设备。它还在云端运行，并提供单一的記憶、上下文和个性化图谱。该代理还可以创建子代理，或面向特定任务的代理，来执行任务。

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 agent can use enterprise business context for knowledge work. However, questions about cost and integration with other tools could be difficult for enterprises.
Google today introduced Gemini Agent, its first workplace personal agent. But enterprises -- even those already in the tech giant’s ecosystem -- will need to ask themselves whether they are ready for AI agents to access their data and context, and whether agents can handle the everyday processes and permissions that employees are used to managing.
Gemini Agent is provisioned with an enterprise’s business context and can use that information for knowledge work, answering questions, creating content and coding -- all from a single prompt box.
Google released the agent during its Gemini at Work 2026 conference in Mountain View, along with a new workflow integration that operates directly inside Google Workspace and connects to its widely used enterprise productivity apps, such Docs, Drive and Sheets
The vendor is the first frontier AI provider to release an agent that connects to enterprise workflows, illustrating how the cloud provider is differentiating itself in the AI race.
“This is a play on digital workspace,” said Mark Beccue, an analyst at Omdia, a division of Informa TechTarget. “It’s an everyday use of a personal assistant in your everyday work.”
While Google is the first frontier AI developer with a workplace personal agent, rivals such as Meta and Microsoft have launched their own digital workers. Meta introduced the Muse personal agent last month; it carries out long-running tasks for consumers and could appeal to small businesses. Microsoft, meanwhile, introduced Copilot Autopilot, an AI digital teammate for enterprise teams.
“It’s the lowest hanging fruit to go after digital workspace,” Beccue said. “It’s really the actualization of an AI interface. We’re going to tell AI what we want to do, and it’s going to do it. That’s where we’re headed.”
However, by going in this direction, Google, Microsoft, and even Meta are moving toward a world in which AI agents need access to enterprise data. And given the recent anxiety about AI agents going out of control, some enterprises might be skeptical about permitting Gemini or other models to access to that data.
However, David Nicholson, an analyst at Futurum Group, said enterprises have to set aside some of their skepticism to realize the power of AI.
“We can’t unlock the real value of what AI can do for an enterprise without the enterprise opening up the safe and letting the agents get in and dig around and have access,” he said.
He added that while some enterprises might pilot Gemini Agent by allowing it access only to some aspects of their organizations, others might go all in. And likely, enterprises that are already using Google Workspace will have no problem allowing the Gemini Agent to access employees’ Gmail, Google Slides and Calendar apps.
“All of that provides the foundational context that seems to be table stakes that you need to expose for certain kinds of operations,” Nicholson said.
“The next question, though, is ‘what about my stuff that has nothing to do with Google?’” he added, noting that enterprises also need to consider data residing in SAP, Salesforce, Oracle, or other big enterprise environments.
Moreover, even if enterprises allow Gemini Agent access to these other environments, a question of permissions and access arises. If the agent tries to access SAP or Microsoft, for example, would it have permission to proceed? Enterprise workers on Microsoft platforms sometimes are restricted from accessing Google files without specific approval.
“It’s not interoperability between what Google Docs does; it's permissions by a company,” Beccue said. “That’s a general theme, that you could say security is costing us productivity. How are agents going to navigate that? Because these are permissions by organizations, not vendors.”
Another challenge is cost, Beccue said. Google said enterprises can choose between Gemini models and Anthropic Claude when using Gemini Agent. However, some of the models are expensive to run. Moreover, having users choose the model they use would require a high level of management and knowledge of token-cost finance, Beccue said.
“It means that we're learning how to be agent prompters or assistant prompters. Now … everybody's responsible for finance?” he continued. “How are organizations going to handle all of this?”
Challenges like these expose that “when you’re starting to deploy workflow agents, you’re not going to be successful on the very first go around, especially in the complex multiapplication environment, where this agent is going to be bridging the gap,” said Sid Nag, founder and analyst at Tekonyx. “To achieve the level of exactness, it’s going to take some time.”
While organizations already in a Google environment might choose Gemini Agent, others might choose to stay closer to their data, Nicholson said.
“It’s the idea of maybe you want to own all of that intellectual property as a separate thing that’s decoupled from one of the big players that might lock you in,” he said.
Despite all these questions about ownership and permissions, Google’s release is a step in the right direction, Nag said.
“It’s a great offering and a great move by Google, integrating their frontier capability with an agentic capability that’s actionable,” he said. It's an overlay middleware API,” he said.
Google said Gemini Agent can be accessed from any device, including web, IOS and Android-based mobile devices. It also runs in the cloud and provides a single set of memories, context and a personalization graph. The agent can also create subagents, or job-specific agents, to do tasks.