9月23日，腾讯WorkBuddy企业版发布办公套件化升级。腾讯文档、腾讯乐享、腾讯网盘、安全与效能运营平台、腾讯AI外观智能体Ardot首批接入。升级后，这些机型以WorkBuddy作为统一的Agent Harness，共用同一个账号、同一份数据和记忆、同一个后台。
这不是一次常规的机型迭代。过去两年，几乎所有办公软件都在机型里嵌入了AI助手，但企业拿到的不是一整套AI赋能的办公平台，而是N套账号、N份记忆、N个后台。文档里的数据、知识库里的资料、网盘里存的文件，这些企业资料就在那里，但没有一个AI能看到全部。
碎片化，正在成为阻碍AI提效的最大门槛。而这一次，腾讯试图从底层应对这个难题。
01
企业AI的真实困境
工作日，一个员工在写项目思路。他先打开文件夹翻出之前的项目记录，切到AI对话框把素材一段段粘贴进去，等AI生成初稿后发现数据对不上，又回到对话框重新交代一遍背景，敲定初稿后又想起网盘里有新素材，又要再跟AI说一遍……
这个情形，正在无数企业里上演。如果从2022年ChatGPT出圈来算，「AI纪元」已经进入第四年了。模型的能力和参数越来越强，但企业的AI落地仍旧步履蹒跚。
麦肯锡今年9月发布的《全球AI应用现状调研》，覆盖了97个市面、逾1700名商业人士，结果表明：近9成企业至少在一项业务职能中常态化使用AI，但真正算得上「AI高绩效企业」的，仅6%。德勤中国的报告给出了更具体的画面：64.4%的企业已进入AI生产部署阶段，但48.5%的企业反馈AI项目实际投资回报低于预期。
一言以蔽之，企业用AI的意愿很强，但最终成效仍旧偏弱。
腾讯云副总裁、CodeBuddy & WorkBuddy负责人刘毅在2026腾讯云AI产业应用大会上说过一句话：「N个超级个体不会自动变成超级团队。」他认为AI时代的团队协同面临三个挑战：缺少人机协作空间、上下文不共享不透明、资产交换成本高。每个软件里塞进一个AI助手，单看都挺聪明，但这些AI彼此不知道对方在做什么。一个员工用AI把8小时的工作压缩到4小时，不等于企业立刻取得翻倍的人效。
从个人提效工具到企业级AI基础设施，中间隔着一段巨大的空白。瓶颈不在模型，在整合。腾讯填上这段空白的方式，就是「套件化」。
02
套件化，腾讯的新解法
所谓套件化，就是让几个软件共用同一个账号、同一份数据和记忆，以及同一个后台。企业买到的是一整套智能体矩阵。文档里记的事、知识库里沉淀的规范、网盘里放的文件，不需要人手搬运到AI面前。AI本来就在里面。WorkBuddy在这里扮演的角色，类似于手机操作平台：底层能力由平台统一封装，各应用调用即可，不必各自从零搭建。
具体来说，这次升级达成了五个统一：账号、积分、记忆与数据、权限与后台、入口和感受。
最基础的是账号打通。五个机型一次接入，员工不用再记五套账号，管理员不用再配五套组织架构。在统一账号基础上，也做到了统一积分池，AI的花费集中管理，花在哪里、花了多少一目了然，企业的AI投入和产出第一次可以核算。权限和数据也跟着统一，文件的智能管理与检索沿用企业原有设置，数据始终留在企业内，人员、时间、交互内容、数据对象与风险行为留在同一条链路，风险可控、可管、可追溯。
这五个统一应对的核心难题，是把企业上下文的获取成本从「人工搬运」降到「平台自带」。其中真正应对上下文断裂的，是统一记忆层与数据层。
那个写思路的员工，反复在文档和对话框之间搬运素材，原因就是AI看不到文档里的内容。升级之后，会议纪要、思路与规范进入腾讯文档，会自动转成Agent能直接读取的上下文。网盘里的文件沿用企业既有权限就能被智能检索。成员在文档内即可一键交给WorkBuddy继续，文件与背景一并带过去。每一次修改与执行沉淀为企业知识，下一轮任务带着上次的结论开工。交付质量不再依赖单次「开盲盒」，而是有了稳定下限。
腾讯不是唯一看到这个方向的企业。也是在9月，微软推出了Microsoft IQ智慧层，让AI智能体基于完整的组织信息开展推理和行动，同时附带权限管理、审计和治理能力。字节则在同月推出「豆包工作」智能体工作平台，与飞书深度融合，共用同一套账号体系和数据权限，为企业给出一站式办公平台。
各家路径不同，方向一致：从「每个软件各长一个AI」转向「共用一个底座」。当主流模型的能力逐步接近，真正拉开差距的，是办公智能体对企业上下文的理解能力。谁能更好地沉淀使用者记忆、贯通任务状态、连接工具与情形，谁就能让AI从「偶尔聪明的助手」变成「可持续工作的生产力平台」。
03
从个人提效到组织资产
上下文打通之后，企业内部发生了哪些变化？
晶科能源给出了一个观察样本。这家头部光伏制造企业今年引入WorkBuddy，进入财务、IT、知识产权、生产等多个业务情形。财务BP Jenny每月做预算分析，需要从多个平台导出数据。不同平台导出的数据格式不统一，过去需要人工一张张整理成统一底稿。第一次做预算分析要两天，每月更新也要两三个小时。这正是上下文断裂的典型表现：数据散落在不同平台里，AI看不到，只能靠人搬运。
引入WorkBuddy后，Jenny把数据整理的规则和方法整理好，让AI按照既定流程完成，在掌握充足资料和背景信息后，这项工作时间从两三个小时压缩到15分钟。Jenny说现在的AI就像「一个知识库很强的应届生」，「告诉它业务背景和规则，它就越来越知道该怎么干」。
上海家化也有类似的例子。这家日化龙头企业在九大部门规模化落地了WorkBuddy企业版，总体平均提效超过4.5倍。但更值得注意的是提效的路径：员工把成熟的工作流程沉淀为Skill，一个部门跑通的能力，其他部门拿来就能用。供应链物流对账、品牌营销等情形中，个人摸索出来的方法和规则，变成了团队内部可以直接调用的能力。
这就是套件化带来的结构性变化。过去，一个团队里有人跑通了AI工作流，换个人用同样的指令，成效就大打折扣。原因很简单：上下文没有沉淀。一个员工工作时的业务背景、历史经验、客户偏好，AI看不到，其他人也无法使用。统一记忆层和数据层打通之后，这些信息留在企业空间里，下一次任务带着上一次的结果直接开工。
管理者的视角也在变。过去面对AI，管理者看到的是若干割裂的报表，每个机型有自己的后台、自己的用量数据、自己的权限设置。统一账号和统一后台之后，管理者第一次能在一个入口看清人、权限、用量与风险。
一个飞轮由此转动：调用越多，企业空间里的上下文越厚；上下文越厚，智能体的交付结果越贴近企业的标准；交付越准，成员越愿意把更多工作交给它。个人的日常提效，逐步积累成团队层面可以共享的智能资产。
结语
套件化是企业AI的「基础设施化」。把上下文、权限、记忆从「每个应用自建」变成「平台统一给出」，腾讯、微软、字节都在走这条路，说明这已经是AI办公的大势所趋。
腾讯在这条路上有自己的先天长处。它具备一个高频使用的办公情形网络，使用者日常交互产生大量真实情形数据，应用间打通多元上下文，推动AI交付成效持续提高。工程能力长期偏向高并发、低延迟、稳定服务，服务超大规模C端使用者的经验迁移到AI时代，让Agent能大规模、多轮、连续地稳定运行。从微信到腾讯会议，腾讯熟悉如何降低使用门槛、提高使用者粘性，而这恰恰是AI Agent进入长期工作流最需要的东西。
当上下文不再断裂，AI共享同一份企业记忆，组织的形态也在被重新定义。晶科能源已经在IT部门试点「一人项目组」，用员工+AI的方式替代过去的臃肿组织，这场变化才刚刚开始，而它最终会把企业带向什么形态，值得每一个管理者和员工深思。
本文来源：https://so.html5.qq.com/page/real/search_news?docid=70000021_9596abc051e05352
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
Enterprise AI's Fragmentation: Tencent Takes Action
SlowDown
2 hours ago
On September 23, Tencent WorkBuddy Enterprise Edition released a suite-based upgrade. Tencent Docs, Tencent Leshang, Tencent Netdisk, the Security & Efficiency Operations Platform, and Tencent AI Visual Agent Ardot were among the first to integrate. Post-upgrade, these products use WorkBuddy as a unified Agent Harness, sharing the same account, the same data and memory, and the same backend.
This is not a routine product iteration. Over the past two years, virtually every office software suite has embedded an AI assistant, yet enterprises have not received a fully AI-enabled office platform; instead, they are left with N accounts, N memory stores, and N backends. Data in documents, materials in knowledge bases, and files in net disks—this enterprise data sits there, but no single AI can see it all.
Fragmentation is becoming the biggest hurdle to AI-driven efficiency. This time, Tencent is attempting to address this challenge from the bottom up.
01
The Real-World Dilemma of Enterprise AI
On a workday, an employee is writing out a project plan. They first open a folder to dig up previous project records, switch to the AI chat box to paste materials in chunks, and after waiting for the AI to generate a first draft, find that the data doesn't match. They have to go back to the chat box to re-explain the background, and after finalizing the draft, remember that new materials are in the net disk, requiring yet another round of briefing the AI...
This scenario is playing out in countless enterprises. If we count from ChatGPT's breakthrough in 2022, the "AI Era" has already entered its fourth year. Model capabilities and parameters are getting stronger, but enterprise AI adoption remains stumbling.
McKinsey's "Global AI Adoption Survey," released this September, covered 97 markets and over 1,700 business professionals. The results showed that while nearly 90% of enterprises regularly use AI in at least one business function, only 6% qualify as "high-performing AI enterprises." Deloitte China's report provided a more specific picture: 64.4% of enterprises have entered the AI production deployment stage, but 48.5% reported that the actual return on investment for AI projects fell short of expectations.
In short, the willingness of enterprises to use AI is strong, but the final results remain weak.
Liu Yi, Vice President of Tencent Cloud and Head of CodeBuddy & WorkBuddy, said at the 2026 Tencent Cloud AI Industry Application Conference: "N super-individuals do not automatically become a super team." He argued that team collaboration in the AI era faces three challenges: a lack of a human-machine collaboration space, lack of shared and transparent context, and high costs of asset exchange. Stuffing an AI assistant into every piece of software may make each one seem smart, but these AIs don't know what the others are doing. One employee compressing an 8-hour task to 4 hours using AI does not equate to an immediate doubling of enterprise productivity.
From personal efficiency tools to enterprise-grade AI infrastructure lies a vast gap. The bottleneck is not the model, but integration. Tencent's way of filling this gap is "suitization."
02
Suitization: Tencent's New Solution
"Suitization" means allowing several software products to share the same account, the same data and memory, and the same backend. What enterprises buy is a complete matrix of agents. Matters recorded in documents, norms accumulated in knowledge bases, and files stored in net disks—there is no need to manually carry them to the AI. The AI is already inside. WorkBuddy plays a role here similar to a mobile operating system: underlying capabilities are uniformly encapsulated by the platform, and apps simply call them, without each having to build from scratch.
Specifically, this upgrade achieved five unifications: account, credits, memory and data, permissions and backend, and entry point and user experience.
The most basic is account integration. With five products integrated at once, employees no longer need to remember five sets of accounts, and administrators no longer need to configure five organizational structures. Based on a unified account, a unified credit pool has also been achieved. AI spending is managed centrally; where it is spent and how much is clear at a glance. For the first time, enterprise AI input and output can be accounted for. Permissions and data are also unified; intelligent management and retrieval of files follow the enterprise's existing settings, ensuring data always remains within the enterprise. Personnel, time, interaction content, data objects, and risk behaviors are kept on the same chain, making risks controllable, manageable, and traceable.
The core challenge addressed by these five unifications is reducing the cost of acquiring enterprise context from "manual transport" to "platform-native." The elements truly addressing context breaks are the unified memory layer and data layer.
That employee writing plans was repeatedly shuttling materials between documents and the chat box because the AI could not see the content in the documents. After the upgrade, meeting minutes, plans, and norms entered into Tencent Docs are automatically converted into context that the Agent can read directly. Files in the net disk can be intelligently retrieved using the enterprise's existing permissions. Members can hand work off to WorkBuddy with one click within the document, carrying over files and background information together. Every modification and execution is deposited as enterprise knowledge, and the next task starts with the previous conclusion. Delivery quality no longer relies on a single "unboxing" experience but has a stable baseline.
Tencent is not the only company seeing this direction. Also in September, Microsoft launched Microsoft IQ Intelligence Layer, enabling AI agents to reason and act based on complete organizational information, while providing permission management, audit, and governance capabilities. ByteDance launched its "Doubao Work" agent work platform in the same month, deeply integrating with Lark to share the same account system and data permissions, providing enterprises with a one-stop office platform.
Different paths, consistent direction: shifting from "each software growing its own AI" to "sharing a common foundation." As mainstream model capabilities gradually converge, what truly creates the gap is the office agent's ability to understand enterprise context. Whoever can better accumulate user memory, connect task states, and link tools and scenarios will turn AI from an "occasionally smart assistant" into a "sustainable productivity platform."
03
From Personal Efficiency to Organizational Assets
After context is connected, what changes occur within the enterprise?
Jinko Solar provided an observation sample. This leading photovoltaic manufacturing enterprise introduced WorkBuddy this year, entering multiple business scenarios such as finance, IT, intellectual property, and production. Finance BP Jenny conducts monthly budget analysis, requiring data export from multiple platforms. The exported data formats are inconsistent, and in the past, it required manual organization into a unified worksheet. The first budget analysis took two days, and monthly updates took two to three hours. This is a typical manifestation of context breaks: data scattered across different platforms, invisible to AI, requiring human transport.
After introducing WorkBuddy, Jenny organized the rules and methods for data sorting and allowed the AI to complete it according to the established process. With sufficient materials and background information, the time for this task was compressed from two to three hours to 15 minutes. Jenny said the current AI is like "a new graduate with a strong knowledge base," and "tell it the business background and rules, and it will increasingly know how to work."
Shanghai Jahwa also has similar examples. This leading daily chemicals enterprise has deployed WorkBuddy Enterprise Edition at scale across nine departments, with an average overall efficiency gain of over 4.5 times. But what is more notable is the path of efficiency: employees distill mature workflows into Skills; capabilities tested in one department can be used by others immediately. In scenarios such as supply chain logistics reconciliation and brand marketing, methods and rules explored by individuals have become capabilities that teams can directly call upon.
This is the structural change brought by suitization. In the past, if one person in a team figured out an AI workflow, another person using the same instructions would see greatly diminished results. The reason is simple: context was not accumulated. The business background, historical experience, and customer preferences of an employee while working were invisible to the AI and unusable by others. Once the unified memory and data layers are connected, this information remains in the enterprise space, and the next task starts directly with the previous results.
The manager's perspective is also changing. In the past, when facing AI, managers saw fragmented reports; each product had its own backend, usage data, and permission settings. After unifying accounts and backends, managers can for the first time see people, permissions, usage, and risks clearly from a single entry point.
A flywheel thus turns: the more it is called, the thicker the context in the enterprise space; the thicker the context, the closer the agent's delivery results are to enterprise standards; the more accurate the delivery, the more willing members are to hand over more work to it. Personal daily efficiency gains gradually accumulate into smart assets shareable at the team level.
Conclusion
Suitization is the "infrastructure" of enterprise AI. Changing context, permissions, and memory from "built by each application" to "provided uniformly by the platform"—Tencent, Microsoft, and ByteDance are all walking this path, indicating that this is the trend for AI office work.
Tencent has its own innate strengths on this path. It possesses a high-frequency office scenario network; users' daily interactions generate massive amounts of real-scenario data, and inter-app integration of diverse contexts drives continuous improvement in AI delivery results. Engineering capabilities have long been biased towards high concurrency, low latency, and stable services; the experience of serving a massive C-end user base transfers to the AI era, enabling Agents to run stably at large scale, over multiple rounds, and continuously. From WeChat to Tencent Meeting, Tencent knows how to lower usage barriers and increase user stickiness, which is exactly what is most needed for AI Agents to enter long-term workflows.
When context is no longer broken and AI shares the same enterprise memory, the form of the organization is being redefined. Jinko Solar has already piloted the "one-person project team" in its IT department, replacing past bloated organizations with a "employee + AI" model. This change has only just begun, and what form it will ultimately take enterprises to is worth deep reflection for every manager and employee.