🔶 Gemini Agent

Google Workspace 中的 Gemini

Gemini 不仅能够回答问题、处理知识型工作并编写代码——它还可直接在 Gmail、Drive、Docs、Slides、Sheets、Chat 和 Calendar 中工作，并继承其在其他场景中具备的记忆、技能和控制能力。它以内联方式工作：位于电子邮件线程中、文档中以及聊天空间中。

在 Workspace 内，Gemini 通过三种方式工作：

• 个人协助：Gemini 可作为你的个人 Agent，在你开始工作前便已了解你的日历、团队、项目，以及文档之间的关联。例如，你可以要求 Gemini 在下周与常规区域活动负责人团队安排会议，而无需提供任何姓名或电子邮件地址。Gemini 会根据你的聊天空间成员以及上次活动的线程识别这些人员，检查他们的日历，并发起电子邮件线程以协调一个适合所有人的时间，即使其中包括外部参与者。它也会以同样方式处理跨应用工作：研究市场趋势、在 Sheets 中构建财务模型，然后创建一份同时呈现两者内容的演示文稿，而你无需在每一步重复说明项目背景。

• 主动委派：Gemini 利用同样的智能能力，主动建议你可委派给它的任务。例如，如果你的经理发送电子邮件，要求你提供最新项目更新的幻灯片演示文稿，Workspace Intelligence 会将其识别为可委派任务，并提供一键交给 Gemini 处理的选项。它也会将相同的推理应用于你的收件箱，呈现最重要而非最新收到的邮件，并解释原因。

• 团队成员：你还可以使用 Gemini 创建与整个团队协作的同事 Agent。你只需描述所需的角色，Gemini 就会创建它。该 Agent 将获得自己的 Workspace 账号，包括电子邮件地址、日历、Drive 和公司目录中的状态信息。你的同事可像与其他任何人协作一样与它合作：将其添加到 Chat 空间中，或通过 @mention 提及它。例如，一位营销经理可以在聊天群组中要求活动协调 Agent 起草一份发布准备文档；任务完成后，Agent 会将文档发布回群组。营销经理也可以在文档评论中标记该 Agent；它可在 Doc 中建议修改并在评论线程中回复，在版本历史记录中以自己的名称出现。同事 Agent 以自身身份而非你的身份行动，并且只能看到你与它共享的内容。访问控制遵循你的团队现有共享和成员资格设置，且不会有任何外部连接器持有你的数据。

数据分析、数据工程与机器学习

我们还将为 Gemini 提供针对特定领域构建的技能，首先从数据领域开始。

数据与分析是一个基础领域，Gemini 可在其中改变业务工作流，使组织能够在数分钟内从自然语言问题获得可执行的运营洞察。我们正在推出专为技术团队和日常业务用户量身打造的能力：

• **面向数据与 ML 工程师：**我们正在通过机器学习技能和工具扩展 Gemini，使每位数据科学家和数据工程师都能获得一支 Agent 团队的支持。工程师可用自然语言描述希望获得的结果，Gemini 随后会生成 PySpark 代码，提供可供编辑和测试的 notebook，训练模型，并自行排查和修复管道问题。

• **面向业务用户：**组织中的任何人只需提出问题，便可生成可靠、实时的运营报告。Gemini 使用与 BigQuery 和 Knowledge Catalog 集成的运营报告技能来构建并保存查询。保存后，团队可按需运行报告而无需产生 token 成本，并保证每次都获得一致、经过验证的答案。

三项 Google Cloud 能力使这些答案有据可依：

• **Knowledge Catalog：**Gemini 可直接集成你的 Knowledge Catalog，一次性映射业务定义，确保所有 Agent 都使用这些定义。它对组织架构模式以及“净利润率”和“可服务市场”等术语相关业务规则的深度理解，可显著提高准确性。无论指标位于 Databricks、dbt、LookML 还是 SAP 中，Gemini 都会直接在其所在位置读取它们，以确保事实性回答。

• **Smart Storage：**90% 的企业数据是非结构化数据——PDF、图像、扫描件和音频记录——它们通常仍处于“暗数据”状态，难以读取。Smart Storage 通过在原地丰富非结构化对象，并将上下文直接写回对象本身来改变这一局面。智能能力保留在字节数据所在之处，并继承你现有的安全态势。

• **无边界 Lakehouse：**组织不应为了使用 AI 而被迫迁移或复制数据。我们的无边界 Lakehouse 允许 Gemini 查询 Amazon S3 和 Azure Data Lake，且不收取浮动出口费用；可直接读取 Salesforce Data 360、SAP 和 Workday 中的数据而无需复制；还能跨 Databricks Unity、Snowflake Horizon 和 AWS Glue 联邦查询开放的 Apache Iceberg 表。

你可以先通过 Knowledge Catalog 注册在 lakehouse 中发现的数据集，然后为 Gemini 分配简单或复杂的分析任务。它会从 Knowledge Catalog 中识别所需数据集，生成计算结果所需的 SQL、Spark 或 Python 代码，并在搭载 Lightning Engine 的 Managed Spark Service 或 BigQuery 中运行代码。它还可以生成图表，并在分析师已经使用的工具中构建仪表板。

通过以 Knowledge Catalog 为数据 Agent 提供基础，Bloomberg Media 在初始开发期间将 SQL 查询准确率提升了 63%。正如 Bloomberg Media 首席技术官 William Anderson 所言：“……通过将我们的 AI 建立在可信的机构上下文之上，我们确保了每一项生成洞察在准确性和质量上的可信度。”

其他客户也从我们的数据解决方案中看到了巨大价值：

• Deutsche Telekom 使用 BigQuery、Cloud Storage 和 Knowledge Catalog 构建统一数据 lakehouse，解决了分布在 40 多个传统本地系统上的碎片化且缓慢的数据基础设施问题，使其团队如今能够以快 10 倍的速度开展工作，同时满足欧洲的数据主权和安全要求。

• Etsy 将 3 PB 数据迁移至统一 lakehouse。通过使用 BigQuery 和开源 Apache Iceberg，他们如今可将关键数据联接的执行速度提升至多 60%，从而能够在更大的数据集上训练模型，并更好地帮助买家匹配理想商品。

• Snap 将其 Prism Agent 直接连接至存储归档，将技术诊断排障时间从 30 分钟缩短至仅 30 秒。

行业特定专业化能力

每个行业都依赖通用模型并不具备的知识。我们正在针对各个行业专门化 Gemini，为其提供各行业特有的技能、工具和知识。它现已面向金融服务和法律行业提供预览，并将很快扩展至政府、医疗保健和零售行业。

• **金融服务：**Gemini for Financial Services 可自动化通常需要跨越多个工作流才能完成的事务：投资研究、信用分析和风险建模。它利用来自 FactSet、LSEG、S&P Global、SEC 备案文件以及你自有专有资料库的可信金融数据。该产品构建于 50 多项基础技能之上，并专为满足严格合规要求而设计；它通过置信度评分、明确的方法论、便于审计的完整数据血缘以及可核查的来源引用展示其推理过程。Gemini for Financial Services 已被 CME Group 和 Deutsche Bank 等全球金融机构使用。

• **法律：**Gemini 可直接继承 NetDocuments 和 iManage 等文档管理平台中基于事项级别的权限和伦理隔离墙。Harvey 等合作伙伴可在满足律所保密标准的同时自动化复杂法律工作，而 Onit 则可在不暴露特权信息的情况下简化合同生命周期管理。Cooley 正在基于 Gemini Enterprise for Legal 构建一款机密信息脱敏 Agent，可在法律文件公开提交前自动完成脱敏处理，包括个人身份信息和机密材料。在高风险诉讼中，这可在保持律所严格客户保密标准的前提下，减少数天的文档审查工作。

此外，我们还看到生态合作伙伴和客户创建了新的垂直应用。例如，Deloitte 创建了一套多 Agent 解决方案，帮助组织识别最具价值的 AI 业务应用机会。其 Agentic AI 蓝图可构建运营热力图，定位适合进行 Agent 化重构的领域，加速部署定制化 Agent 工作流，并将实现可衡量业务影响的路径最多缩短至原来的三分之一。

保护和治理 Agent

两个因素决定企业 Agent 项目能够成功还是停滞：你能否对其进行治理，以及你能否承担其成本。

在全组织范围内部署自主 Agent 时，治理归结为回答四个基本问题：

• Agent 是谁，它具有什么身份？

• 它被允许做什么，或者被授予了哪些权限？

• 它做了什么，我可以在哪里查看其执行记录？

• 它绝不应接触什么？

身份、策略和可观测性回答了前三个问题。第四个问题由 Agent Gateway 解决。

• **身份：**每个 Agent 都拥有自己的身份，该身份经过密码学证明，并像员工一样受到治理，具有最小权限访问控制。该身份会写入记录其工作的日志，也会写入为代表它运行代码而启动的任何虚拟机中。

• **授权与权限：**你可以为每个 Agent 授予细粒度、基于角色的访问权限，并由组织的安全管理员批准。当 Gemini Agent 连接至外部系统时，其身份会通过 OAuth 等行业标准进行映射和传播。

• **审计：**此外，Gemini 采取的每项操作都会写入审计轨迹，并归属于 Agent 而非个人。由于该身份会进入 Agent 为运行代码而启动的任何虚拟机，你可以使用我们的可观测性工具实时监控这些日志，并在异常行为产生影响前发现它。

• **策略管理与控制：**Gemini 默认提供身份、发现、治理和安全能力。所有 Gemini Agent 都会在拥有独立网络边界的 Agent Sandbox 中安全执行任务。所有流量——入站、出站以及 Agent 之间的流量——都会经过 Agent Gateway；这是一个 AI 网络防火墙，可实时执行组织策略。你只需编写一次策略，例如：“Agent 不得打开被归类为 Need to Know 的文档。”Gemini 会将其应用于公司内的每个 Agent，而不需要逐一检查每个 Agent。

领先的成本控制与支出选项

尽管自 2024 年以来，每 token 价格已下降 98%，但企业 AI 使用量已呈爆炸式增长。将每个简单循环都交由高端模型运行，会迅速耗尽企业预算。我们最近宣布了新的灵活支出选项，今天还将进一步增加以下能力：

• **多模型编排：**如前所述，Gemini 允许你为工作选择合适的模型，甚至可在大型项目中为特定任务组合不同模型，以实现最佳质量并降低成本。

• **智能路由：**我们的路由工具会自动分流企业工作负载，使每项工作负载均在能够以最低成本提供最高性能的模型上运行。

• **实时支出上限：**你可以在 Cloud Billing Console 中为项目的 AI 支出设置硬性上限，Gemini 会强制执行。Gemini 会监控 token 用量和沙箱成本；一旦触发支出上限，该项目的 Agent 将暂停运行，你可在控制台中一键选择恢复工作。由于追踪粒度为项目级别，企业可以将 AI 成本分摊回具体部门。

基础设施：性价比提升 80%

所有这些创新——Agent、数据、身份以及其他一切——均运行于我们的 AI Infrastructure 和模型之上。它们正是这一经济模式得以成立的原因。

首先是我们的 AI 基础设施。我们将芯片、网络、数据中心以及连接这一切的软件协同设计为一个高度优化的系统——即我们的 AI Hypercomputer。这种垂直整合可带来显著优于通用配置的性价比，而我们仍在持续推进这一领域的技术前沿。例如，我们最新的 TPU 8i 系统相比上一代实现了 80% 更优的性价比，使复杂、高价值的 Agent 任务既快速又极具成本效益。

在这些硅芯片之上运行的是我们的模型：用于前沿推理的 Argon、用于速度与高吞吐的 Flash、用于生成式媒体的 Omni，以及面向轻量级开放权重边缘工作负载的 Gemma。

边缘能力已经成为现实。NASA 喷气推进实验室正在轨道卫星上直接运行 Gemma，这是视觉语言模型的首次实现。JPL 的 NAVI 软件在卫星机载端分析地球影像，并向地面操作人员发送包含关键信息的摘要。

由我们的生态系统进一步增强

我们始终致力于在技术栈的每一层提供选择，确保你可以自由选择最适合的 t

Gemini in Google Workspace

Gemini not only answers your questions, handles your knowledge work, and writes your code — it works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, carrying the same memory, skills, and controls it has everywhere else. It works inline: in the email thread, in the document, and in the chat space.

Inside Workspace, Gemini works three ways:

• Personal assistance : Gemini works as your personal agent and arrives already knowing your calendar, your team, your projects, and how your documents relate to one another. It is briefed before you start. For example, you can ask Gemini to set up a meeting with the usual team of regional event leads next week, without supplying a single name or email address. Gemini determines who those people are from the membership of your chat space and the thread from your last event, checks their calendars, and starts an email thread to coordinate a time that works, even with external participants. It handles work that spans applications the same way: researching market trends, building a financial model in Sheets, then creating a deck that presents both, without you re-explaining the project at each step.

• Proactive delegation : Gemini uses this same intelligence to proactively suggest tasks you can delegate to it. For example, if your manager emails you asking for the latest project update as a slide deck, Workspace Intelligence recognizes that as a delegatable task and gives you a single-click option to pass it to Gemini. It applies the same reasoning to your inbox, surfacing the message that matters most rather than the one that arrived last, and explaining why.

• A member of your team : You can also use Gemini to create a coworker agent that works with your entire team. You simply describe the role you need, and Gemini creates it. The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory. Your colleagues work with it the way they work with anyone else, by adding it to a Chat space or @mentioning it. For example, a marketing manager can ask an events coordinator agent in a chat group to draft a launch readiness document, which it posts back to the group when complete. The marketing manager could also tag the agent in a document comment, where it can suggest an edit in the Doc and reply in the comment thread, appearing under its own name in version history. A coworker agent acts under its own identity rather than yours, and it sees only what you share with it. Access follows the sharing and membership your team already uses, and no outside connector holds your data.

Data analysis, data engineering and machine learning

We are also giving Gemini skills built for specific domains, starting with data.

Data and analytics represent a foundational domain where Gemini transforms business workflows, enabling organizations to move from plain-language questions to actionable operational insights in minutes. We are introducing purpose-built capabilities tailored to both technical teams and everyday business users:

• For Data and ML engineers: We are extending Gemini with machine learning skills and tools that back every data scientist and data engineer with a team of agents. Engineers describe the outcome they want in plain language, and Gemini generates PySpark code, provides notebooks to edit and test it, trains models, and troubleshoots and fixes pipeline issues on its own.

• For business users: Anyone in your organization can generate reliable, real-time operational reports simply by asking. Gemini uses operational reporting skills integrated with BigQuery and the Knowledge Catalog to construct and save the query. Once saved, teams can run reports on demand without incurring token costs, guaranteeing consistent, verified answers every time.

Three Google Cloud capabilities keep those answers grounded:

• Knowledge Catalog: Gemini integrates directly with your Knowledge Catalog to map business definitions once so all agents use them. This deep understanding of your organization’s schemas and business rules for terms like “net margin” and “addressable market” dramatically improves accuracy. Whether your metrics live in Databricks, dbt, LookML, or SAP, Gemini reads them directly where they sit to ensure factual responses.

• Smart Storage : Ninety percent of enterprise data is unstructured — PDFs, images, scans, and audio recordings — which often remain "dark" and unreadable. Smart Storage changes this by enriching unstructured objects in place and writing context back directly onto the object itself. The intelligence stays where the bytes reside and inherits your existing security posture.

• Borderless Lakehouse: Organizations should not be forced to move or duplicate their data to use AI. Our borderless Lakehouse allows Gemini to query Amazon S3 and Azure Data Lake with no variable egress fees, read directly from Salesforce Data 360, SAP, and Workday without copying data, and federate open Apache Iceberg tables across Databricks Unity, Snowflake Horizon, and AWS Glue.

You can start by registering data sets that you discover in your lakehouse with the Knowledge Catalog, and then assign Gemini simple or complex analysis to perform. It identifies the necessary data sets from the Knowledge Catalog, generates the necessary SQL, Spark or Python code to calculate the results, and runs the code on our Managed Spark Service with Lightning Engine or in BigQuery. It can also generate charts and build dashboards in the tools your analysts already use.

By grounding its data agents in the Knowledge Catalog, Bloomberg Media lifted its SQL query accuracy by 63% during initial development. As William Anderson, Bloomberg Media’s CTO, noted: “...by grounding our AI in a trusted institutional context, we ensure confidence in the accuracy and quality of every insight generated.”

Additional customers are seeing tremendous value from our data solutions:

• Deutsche Telekom used BigQuery, Cloud Storage, and Knowledge Catalog to build a unified data lakehouse to overcome a fragmented and slow data infrastructure spread across 40+ legacy on-premise systems, allowing their teams to now move ten times faster while addressing European data sovereignty and security requirements.

• Etsy migrated three petabytes of data to a unified lakehouse. By using BigQuery and open-source Apache Iceberg, they can now perform critical data joins up to 60% faster, allowing them to train models on much larger datasets and better connect buyers with the perfect items.

• Snap connected its Prism agent directly to its storage archives, cutting technical diagnostic troubleshooting times down from 30 minutes to just 30 seconds.

Industry-specific specializations

Every industry runs on knowledge a general model does not have. We are specializing Gemini for individual industries, with the skills, tools, and knowledge specific to each one. It is now in preview for Financial Services and Legal, and coming soon to Government, Healthcare, and Retail.

• Financial Services: Gemini for Financial Services automates what is usually pieced together across many workstreams: investment research, credit analysis, and risk modeling. It draws on trusted financial data from FactSet, LSEG, S&P Global, SEC filings, and your own proprietary repositories. Built with more than 50 foundational skills and designed to satisfy rigorous compliance, it shows its reasoning through confidence scores, explicit methodologies, full data lineage for easy auditing, and source citations you can check. Gemini for Financial Services is already being used by global financial institutions like CME Group andDeutsche Bank.

• Legal: Gemini inherits matter-level permissions and ethical walls directly from document management platforms like NetDocuments and iManage. Partners like Harvey automate complex legal work while meeting the firm's confidentiality standards, while Onit streamlines contract lifecycle management without exposing privilege. Cooley is building a confidential information redaction agent on Gemini Enterprise for Legal that automates redactions, including personally identifiable information and confidential material, before legal filings go public. In high-stakes litigation, this can cut days of document review while maintaining the firm's rigorous standards for client confidentiality.

In addition, we’ve seen our ecosystem and customers create new vertical applications as well. For example, Deloitte has created a multi-agentic solution to help organizations identify the highest-value opportunities for AI applications for their business. Their Agentic AI blueprint develops operational heat maps to pinpoint areas ready for agentic redesign, accelerating the deployment of custom agentic workflows and achieving up to a 3x faster path to measurable business impact.

Securing and governing agents

Two factors determine whether an enterprise agent program succeeds or stalls: whether you can govern it, and whether you can afford it.

When deploying autonomous agents across an organization, governance comes down to answering four fundamental questions:

• Who is the agent, and what identity does it have?

• What is it allowed to do or what permissions is it granted?

• What did it do and where can I see what it did?

• What should it never touch?

Identity, policy, and observability answer the first three. The fourth is Agent Gateway.

• Identity: Every agent gets its own identity, cryptographically attested and governed like an employee, with least-privilege permissions. That identity is stamped into the logs that capture its work, and into any virtual machine spun up to run code on its behalf.

• Authorization and permissions: You give each agent fine-grained, role-based access permissions, approved by your organization’s security administrators. When the Gemini agent connects to an external system, its identity is mapped and propagated through industry standards such as OAuth.

• Auditing: In addition, every action Gemini takes is written to an audit trail and attributed to the agent rather than to a person. Since the identity travels into any virtual machine the agent spins up to run code, you can monitor those logs with our observability tools in real time and catch anomalous behavior before it matters.

• Policy management and control: Gemini gives you identity, discovery, governance, and security by default. All Gemini agents execute tasks safely inside an Agent Sandbox with its own network boundary. All traffic — in, out, and between agents — passes through Agent Gateway, an AI network firewall enforcing your organization’s policies in real time. You write the policy once, such as: “agents may not open documents classified Need to Know.” Gemini applies it to every agent in your company instead of checking one agent at a time.

Leading cost controls and spending options

While per-token prices have dropped 98% since 2024, enterprise AI volume has exploded. Running every simple loop through a premium model quickly breaks corporate budgets. We announced new flexible spending options recently and are adding to those today:

• Multi-model orchestration: As mentioned earlier, Gemini allows you to choose the right model for the job, and even combine different models for specific tasks in larger projects to deliver optimal quality and lower your cost.

• Smart routing: Our routing tool automatically triages enterprise workloads so each one runs on the model that delivers maximum performance at the lowest possible cost.

• Real-time spend caps: You can**** set a hard limit on a project's AI spend in the Cloud Billing Console, and Gemini enforces it. Gemini monitors token usage and sandbox costs, and if a spend cap is triggered, that project's agent pauses, and you can choose to resume work with a single click in the console. Because the tracking is per project, companies can charge AI costs back to specific departments.

Infrastructure: 80% better price-performance

All of this innovation — the agent, the data, the identity, and everything else — runs on our AI Infrastructure and our models. They are the reason the economics work.

First, our AI infrastructure. We co-design the chips, the network, the data center, and the software that connects it all as one highly-optimized system — our AI Hypercomputer. This vertical integration delivers significantly better price-performance than the generic setups, and we are continuing to push the state of the art here. For example, our latest TPU 8i system delivers 80% better price-performance than the prior generation, making complex, high-value agent tasks both fast and highly affordable.

On top of this silicon sit our models: Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for lightweight, open-weights edge workloads.

The edge capabilities are already real. NASA’s Jet Propulsion Laboratory is running Gemma directly on a satellite in orbit, a first for a vision-language model. JPL’s NAVI software analyzes Earth imagery on board the satellite and sends a summary with key information to ground operators.

Supercharged by our ecosystem

We remain committed to choice at every layer of the stack, ensuring you have the freedom to choose the best tech stack for your unique needs.

Inside Gemini, you have a broad choice of third-party agents from leading SaaS providers as well as connectors that are ready to use. They are even easier to discover as collections in the new AI Solution Finder.

Around Gemini, our services ecosystem continues to rapidly scale to meet enterprise demand. Together, in the last month alone, our global consulting partners have put more than a hundred thousand of their developers and consultants through Gemini hackathons.

In addition, Accenture has made a significant investment by establishing a dedicated Gemini Enterprise Business Group to accelerate customer adoption and innovation. Several other partners have also expanded their commitments to support the growing demand for Gemini, including establishing specialized centers of excellence and training Gemini credentialed practitioners to help customers build and deploy production-grade agents faster.

Gemini Enterprise in action: Global momentum

Organizations worldwide are driving measurable impact across every major region:

Europe and Middle East

• Arden University , a UK provider of flexible online and blended learning, is preparing students for the global job market by rolling out Gemini Enterprise. Over 30,000 students and staff will now have access to advanced Al tools to personalize their studies and gain essential skills for the workplace.

• Commerzbank , one of Germany’s largest private banks, is using Gemini Enterprise to automate document quality assurance, reducing manual work on document reviews from 20 hours to just one. They are expanding this with a multi-agent system for AI-driven QA.

• Lloyds Banking Group is scaling agentic AI across the bank. Envoy, a secure, Gemini-powered pipeline, has already onboarded nearly a thousand engineers building production-ready AI use cases. From fraud detection to commercial client onboarding, Envoy gives teams a reusable agent marketplace, a "golden path" for building agents, and a CI/CD pipeline — all wrapped in the guardrails and security a highly regulated bank demands.