
WorkBuddy还在提速。
9月23日,腾讯WorkBuddy企业版发布办公套件化升级最新进展,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五款机型首批接入。
升级后,WorkBuddy企业版售价保持不变,企业一次购入,就可以把统一账号、统一积分池、统一记忆层与数据层,和五大机型的会员权限一次收齐。
例如Ardot给出外观席位,在WorkBuddy“外观创意模式”里生成的外观初稿,可以直接在Ardot里精修、协作、交付。
部分机型给出更高版本的权益,可以在企业控制台按需增购。
从单个应用,到一套平台
这次升级的核心价值,就是把WorkBuddy从“能调用工具的AI入口”,升级为“企业办公Agent的操作平台底座”。它借助打通办公套件的上下文与账号体系、管理后台,完成腾讯在企业级AI办公的一次战略卡位。
需要补充的是,套件化不是把几个软件打包一起卖。它在AI办公时代的真实含义是,AI应用共用同一个 AI 底座,涵盖同一个账号、同一份记忆、同一个后台。
总的来说,有几点变化:
接入方式变了。腾讯文档、腾讯乐享、腾讯网盘等机型,不需要再单独打开,直接内化成WorkBuddy可直接调用的能力。
这意味着,员工不用再记五套账号,管理员不用再配五套组织架构,一次开通即可在套件内通行。
角色分工变了。最核心的变化是,WorkBuddy在套件里承担了Agent Harness的角色,它用统一的规划、技能与上下文能力给所有机型供能。 也就是账号体系走One ID,一套身份打通。
账号和账本统一了。统一账号、统一积分池、统一后台。管理员在一个入口,集中管理所有机型的Credit(积分用量)用量。
而且,企业不用再为每个机型的AI能力单独付费、单独开单。钱花在哪里、花了多少一目了然,投入与产出第一次能放在一起核算。
记忆层和数据层打通了。腾讯文档成了公司的长期记忆,会议纪要、思路、规范沉淀在里面,Agent随取随读;乐享做可信知识中枢,把重复、过期、口径冲突的知识治理干净;Ardot接住专业外观的精修和交付;安全与效能运营平台,是这套体系的仪表盘。
基于此,企业的办公内容创作、资产沉淀、再到能力复用,往后都在WorkBuddy里统一完成,形成以企业知识流转、内容生产为核心的一体化智能体矩阵。
文档里记的事、知识库里沉淀的规范、网盘里放的文件,不再需要人手搬运到 AI 面前,因为AI本来就在里面。
例如某ToB思路交付服务商自行测算,联合使用WorkBuddy与腾讯乐享后,在相关交付情形中,人均每天约节省 2.3 小时。
AI办公效率时刻,WorkBuddy要当内核和基座
要知道,企业级AI办公的“效率堵点”早就不是模型能力,而在上下文断裂:员工每打开一个新对话,AI就要重新理解企业背景。
套件化升级统一了记忆层与数据层,使各端AI可以读写同一份办公记忆和数据,Agent不用从零理解任务背景,交付质量也就从依赖单次表现,转为有一个稳定的下限。
以腾讯文档为例。会议纪要、思路、规范进了文档,转成Agent可直接读取的上下文,文档成了企业长期记忆;
成员在文档里一键交给WorkBuddy继续,文件与背景一并带过去;每次修改与执行都沉淀为企业知识,下一轮任务带着上次的结论开工。
其带来的收益是多维度的:
员工这边,人机共写一份文件,指令写进原文,改动自动合并、可追溯,下一轮不用从头再来;他们跑通的流程还能固化成 Skill,一个人的经验变成团队能直接复用的能力。
对管理者来说,其面对的是完整业务链路,不是割裂的报表;一个后台,完成人员、权限、用量的分配核查,风险行为留在同一条链路上,同样可管、可查、可追溯。
另外,IT与安全的边界也不必重建。账号、权限、数据沿用原有设置,老规矩,不用推倒重来。对企业本身,经验变成了资产。知识持续沉淀,个人经验长成组织能力,就算员工离职,他的经验也留下来了。
当干活的人和Agent,被串联在同一个圈里,企业AI提效的飞轮就转起来了。
例如日化龙头上海家化,它已在九大部门规模化落地WorkBuddy企业版,总体平均提效超过4.5倍。在供应链物流对账、品牌营销等情形中,员工把成熟流程沉淀为Skill之后,一个部门跑通的能力,其他部门可直接复用。
家化公司反馈称,WorkBuddy已不只是个体提效工具,正在逐步变成嵌入业务流程的“数字员工”。
不难看出,WorkBuddy想做的不仅是工具,而是做一个能承载所有工具的平台,让企业用AI 这件事情掌握更多的自主权。
它挑选的路径是先做好基座,再让应用和生态自己长出来。
一台电脑的操作平台,内核管调度、管资源分配,所有软件都跑在它上面,应用按它的规矩调用能力,这台电脑就是你自己的电脑,而不是别人的。
WorkBuddy做好底座,企业在上面搭建应用、调度资源,这就是企业自己的Agent,而不是别人的。
WorkBuddy想扮演的就是这个角色。
这背后是WorkBuddy一个格外清醒的判断:Agent的实际运行是一个循环——获取上下文、推理与规划、调用工具并采取行动、观察与验证结果。Harness,就是让这个循环稳定跑下去、且跑得好的底座。
并且,WorkBuddy还为构建的Harness定下两个目标:提高Agent首次执行的正确率,给出反馈循环让平台自动发现、纠正常见难题。
友商还在做应用,WorkBuddy已经在做基座
这个判断,正在成为行业共识。
阿里云在今年云栖大会上提出的“Agentic Cloud”战略,核心就是模型、Harness、上下文三件事绑一起。OpenAI在9月发布Agents API,将Codex Harness封装为托管云服务;百度搭子最近也推出Harness智能引擎,负责任务规划、工具调用和执行流程的统筹调度。
在我看来,Harness正在成为AI办公的“操作平台内核”。谁掌握Harness,谁就掌握企业 AI 的入口;谁掌握企业上下文,谁就掌握Harness的护城河。
区别在于,友商还在各自的生态里做应用:豆包工作绑着飞书体系做 AI,千问办公顺着钉钉情形铺智能体,百度搭子攒其个人端工具箱;WorkBuddy已经在搭OS底座,把套件形态落地,把概念做成机型,领跑行业至少大半年。
沙利文发布的《2026年全球桌面AI智能体市面研究报告》数据表明,WorkBuddy同时位列中国个人端桌面智能体第一、中国企业级桌面智能体第一。这个“双第一”,恰好对应WorkBuddy协助企业完成从“超级个体到AI Native超级团队”的进化。
在AI办公进入“效率时刻”这个赛点上,领跑也容易形成马太效应。2026年6月WorkBuddy PC端单月访问量达2097万次,超过第二名和第三名综合,规模长处也带来数据积累和生态吸引力。
过去两年,几乎所有办公软件都在做同一件事,即,在自己的机型里塞一个AI助手。“AI+办公”的结果就是每家企业被迫面对N套账号、N份记忆、N个后台、N张账单。
这次套件化升级把逻辑调了个头,其底色也变成“AI原生办公”。WorkBuddy跑在了前面,也更接近达成Agent操作平台的目标。
这符合腾讯多年的开放逻辑,也就是把已验证的能力变成“公共设施”,让更多人能复用与共创。只是这次,对象从应用升级为智能体。
简单说,WorkBuddy在AI时代继续把这个“传统技能”延续、平台化,并摸索出一个Agent将来最健康的商业模式:
开放带来的不只是接入量,更是上下文密度。开放得越广,上下文越厚,模型、机型和 Agent Harness 的迭代就越快,应用感受也越好,这又引来更多伙伴、开发者接入,最终形成更深的生态护城河,迁移成本也一起抬高。
WorkBuddy也需要正视,友商各有强项。豆包工作强在组织,跟飞书绑得深,协同感受是它的主场;千问办公强在流程,钉钉里跑通的审批和数据流;金山办公沉淀了38年的格式功底,文档编辑能力暂时无人能比。百度搭子增速最猛,个人端工具箱铺得飞快。
更关键的是,套件化越彻底,绑定越深,迁移成本越高,这也会让企业有所权衡,是不是要把全套办公家底都交出去。
更何况,阿里、字节、OpenAI、百度、金山办公都在往Harness、上下文上押重注,WorkBuddy先发窗口期可能也不会太长。
本文来源:https://www.citnews.com.cn/news/222345
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
WorkBuddy is still accelerating.
On September 23, Tencent WorkBuddy Enterprise Edition released its latest progress on office suite upgrades — five products were first integrated: Tencent Docs, Tencent Lexiang, Tencent Pan, the Security & Efficiency Operations Platform, and Tencent AI appearance agent Ardot.
After the upgrade, WorkBuddy Enterprise Edition's price remains unchanged. With one purchase, enterprises get unified accounts, unified credit pools, a unified memory layer and data layer, plus membership entitlements for all five products.
For example, Ardot provides an appearance workstation. Drafts generated in WorkBuddy's "Appearance Creative Mode" can be directly refined, collaborated on, and delivered in Ardot.
Some products offer higher-tier entitlements that can be purchased on demand in the enterprise console.
From a single application to a unified platform
The core value of this upgrade is to elevate WorkBuddy from an "AI entry point that can call tools" to an "operating platform foundation for enterprise office Agents." By connecting the office suite's context, account system, and admin console, it completes Tencent's strategic positioning in enterprise-grade AI office.
It should be added: suite-ification is not just bundling several software products together. In the AI office era, it really means AI applications share the same AI foundation, covering the same account, the same memory, and the same backend.
Overall, there are several changes:
The integration method has changed. Tencent Docs, Tencent Lexiang, Tencent Pan, and other products no longer need to be opened separately; they are directly internalized as capabilities WorkBuddy can call.
This means employees no longer need to memorize five accounts, and admins no longer need to configure five org structures. One activation grants access across the suite.
Roles have changed. The core change is that WorkBuddy plays the role of Agent Harness in the suite, powering all products with unified planning, skills, and context capabilities. The account system runs on One ID — one identity across everything.
Accounts and ledgers are unified. Unified account, unified credit pool, unified backend. Admins centrally manage all products' credit usage from a single entry point.
Moreover, enterprises no longer need to pay separately or raise separate orders for each product's AI capabilities. Where money goes and how much is spent is clear at a glance; for the first time, input and output can be accounted together.
The memory layer and data layer are connected. Tencent Docs becomes the company's long-term memory, where meeting minutes, ideas, and standards settle, and Agents can read them on demand; Lexiang serves as a trusted knowledge hub that cleans up repeated, outdated, or conflicting knowledge; Ardot handles the refinement and delivery of professional appearance work; the Security & Efficiency Operations Platform is the dashboard for the whole system.
Based on this, enterprise office content creation, asset accumulation, and capability reuse will all be completed within WorkBuddy, forming an integrated agent matrix centered on enterprise knowledge flow and content production.
Things recorded in documents, standards settled in the knowledge base, and files stored in the drive no longer need to be manually carried to the AI — because the AI is already inside them.
For example, one B2B delivery service provider's own estimates show that after combining WorkBuddy with Tencent Lexiang, in relevant delivery scenarios, each person saves about 2.3 hours per day.
At the moment of AI office efficiency, WorkBuddy aims to be the core and foundation
It should be noted that the "efficiency bottleneck" of enterprise AI office has long ceased to be model capability — it's context fragmentation: every time an employee opens a new conversation, the AI has to re-understand the enterprise background.
The suite upgrade unifies the memory and data layers, allowing AI on all sides to read and write the same office memory and data. Agents no longer need to understand task backgrounds from scratch, so delivery quality shifts from depending on one-off performance to having a stable floor.
Take Tencent Docs as an example. Meeting minutes, ideas, and standards go into documents and are converted into context Agents can directly read; documents become the enterprise's long-term memory.
Members can hand off to WorkBuddy in one click inside a document, with files and background carried along; every modification and execution settles into enterprise knowledge, and the next round of tasks starts with the previous conclusions.
The benefits are multi-dimensional:
For employees: human and AI co-write one document, instructions are written into the original text, changes are auto-merged and traceable, and the next round doesn't start from scratch; the workflows they run can be solidified into Skills, turning one person's experience into a capability the whole team can directly reuse.
For managers: they face a complete business chain instead of fragmented reports; one console handles the allocation and review of personnel, permissions, and usage, with risk behavior staying on the same chain — manageable, inspectable, and traceable.
In addition, the boundary of IT and security doesn't need to be rebuilt. Accounts, permissions, and data follow the original settings — old rules, no need to overturn and start over. For the enterprise itself, experience becomes an asset. Knowledge keeps settling, personal experience grows into organizational capability, and even if an employee leaves, their experience remains.
When people and Agents are linked in the same loop, the flywheel of enterprise AI efficiency starts spinning.
For example, Shanghai Jahwa, a leading daily chemical company, has already rolled out WorkBuddy Enterprise Edition on a large scale across nine departments, with overall average efficiency improved by over 4.5x. In scenarios such as supply chain logistics reconciliation and brand marketing, after employees settle mature workflows into Skills, the capabilities proven in one department can be directly reused by other departments.
Jahwa reported that WorkBuddy is no longer just an individual efficiency tool — it is gradually becoming a "digital employee" embedded in business processes.
It's easy to see that WorkBuddy wants to be not just a tool, but a platform capable of hosting all tools, giving enterprises more autonomy in using AI.
The path it has chosen is to build a solid foundation first, then let applications and the ecosystem grow on their own.
On a computer's operating system, the kernel handles scheduling and resource allocation. All software runs on top of it, and applications call capabilities according to its rules. That computer is your own computer, not someone else's.
When WorkBuddy builds a solid base, and enterprises build applications and allocate resources on top of it, that is the enterprise's own Agent — not someone else's.
That is the role WorkBuddy wants to play.
Behind this lies WorkBuddy's particularly clear judgment: an Agent's actual operation is a loop — acquiring context, reasoning and planning, calling tools and taking action, observing and verifying results. The Harness is the foundation that keeps this loop running stably and running well.
Moreover, WorkBuddy has set two goals for the Harness it builds: improve the accuracy of an Agent's first execution, and provide a feedback loop that lets the platform automatically discover and fix common problems.
Peers are still building applications; WorkBuddy is already building the foundation
This judgment is becoming industry consensus.
Alibaba Cloud's "Agentic Cloud" strategy at this year's Apsara Conference centers on binding models, Harness, and context together. OpenAI released the Agents API in September, packaging Codex Harness as a managed cloud service; Baidu Dazi recently launched a Harness intelligent engine for the orchestration and dispatch of task planning, tool calling, and execution flows.
In my view, Harness is becoming the "operating system kernel" of AI office. Whoever masters Harness controls the entry point of enterprise AI; whoever masters enterprise context holds the moat of Harness.
The difference: peers are still building applications within their own ecosystems — Doubao Work binds to Feishu to build AI, Qianwen Office follows DingTalk scenarios to deploy agents, Baidu Dazi is building out its personal toolbox; WorkBuddy is already building the OS foundation, turning the suite format into reality, turning concepts into products, leading the industry by at least six months to a year.
According to Frost & Sullivan's "2026 Global Desktop AI Agent Market Research Report," WorkBuddy ranks first in both China's personal desktop agents and China's enterprise desktop agents. This "double first" corresponds exactly to WorkBuddy helping enterprises evolve from "super individuals to AI-native super teams."
At the race point where AI office enters the "efficiency moment," leading can easily create a Matthew effect. In June 2026, WorkBuddy's PC monthly visits reached 20.97 million, surpassing the second and third place combined. Scale advantages also bring data accumulation and ecosystem appeal.
Over the past two years, almost all office software has been doing the same thing: stuffing an AI assistant into their own products. The result of "AI + office" is that every enterprise is forced to deal with N accounts, N memories, N backends, and N bills.
This suite upgrade flips the logic, and its foundation becomes "AI-native office." WorkBuddy is running ahead, and is closer to achieving the goal of an Agent operating platform.
This aligns with Tencent's long-standing logic of openness — turning proven capabilities into "public utilities" so more people can reuse and co-create. Only this time, the subject has upgraded from applications to agents.
In short, WorkBuddy continues this "traditional skill" in the AI era, makes it a platform, and is exploring what may become the healthiest business model for Agents in the future:
Openness brings not just more integrations, but more context density. The broader the openness, the thicker the context, the faster models, products, and the Agent Harness iterate, and the better the user experience. This in turn attracts more partners and developers to integrate, eventually forming a deeper ecosystem moat and raising switching costs.
WorkBuddy also needs to acknowledge that peers each have their strengths. Doubao Work is strong in organization, deeply bound to Feishu, with collaboration experience as its home turf; Qianwen Office is strong in processes, with approval flows and data streams running through DingTalk; Kingsoft Office has 38 years of format expertise, and its document editing capability is currently unmatched. Baidu Dazi is growing the fastest, rolling out its personal toolbox rapidly.
More critically, the more thorough the suite-ification, the deeper the binding, and the higher the switching cost — which also makes enterprises weigh whether to hand over their entire office stack.
Moreover, Alibaba, ByteDance, OpenAI, Baidu, and Kingsoft are all betting heavily on Harness and context. WorkBuddy's first-mover window may not last long.