腾讯WorkBuddy开放平台上线,距离“Agent时代操作系统”还有多远?Tencent WorkBuddy Open Platform Launches — How Close Are We to an “Agent Era Operating System”?


AI free token/AI免费积分

腾讯WorkBuddy开放平台上线,距离“Agent时代操作系统”还有多远?

9月2日,腾讯WorkBuddy正式宣告开放平台上线,首批拉入超百家生态合作伙伴,同时首发9款联名智能硬件,接入30多个行业应用,并向开发者同步开放Skill、Expert、Connector三大核心能力。

发布会上,腾讯云副总裁、腾讯CodeBuddy与WorkBuddy负责人刘毅对WorkBuddy的定位给出了极高的定调——要做"面向Agent时代的操作平台,不是做一个更强的工具,而是做一个能承载所有工具的平台"。

与阿里和字节相比,腾讯动手最早,也是唯一一家将硬件、应用、开发者三个层面同时铺开的厂商。速途网认为,这次开放生态中,WorkBuddy真正押注的是底座和上下文资产,而开放能否成功,归根结底取决于合作伙伴能否在其中获得商业回报。

三个月内三家巨头悉数入场,腾讯是否已经拉开了代际差距?

2026年的办公Agent赛道,直到8月才真正形成格局。

8月3日,阿里巴巴宣布"千问办公"开启公测,该产品由QoderWork、MuleRun、悟空三者整合而成,打通了覆盖超2000万家企业的钉钉;8月25日,字节推出"豆包工作",而在此之前,飞书于7月30日、TRAE与扣子于8月24日已相继整合进豆包体系。

腾讯布局时间最早,WorkBuddy于今年3月面世,领先阿里和字节将近5个月。根据AI机型榜(AICPB)数据,7月WorkBuddy在PC客户端口径下月活达1115.23万,环比增幅304.4%,位列AI办公智能体总榜首位;字节的TraeWork为265.99万,阿里千问办公为60.79万,百度搭子则以674.3万月活夺得增速第一。

然而,WorkBuddy的先发优势似乎并未转化为代际差距。华尔街投行杰富瑞的实测报告显示,千问办公、豆包、WorkBuddy的综合评分分别为95分、77分、66分。这也从侧面表明,WorkBuddy的领先更多依赖于入口优势和先发红利,而非任务执行本身的质量。

因此,笔者看到了腾讯WorkBuddy构建开放生态的举措。在硬件层面,围绕"听、看、记、聊、协"五种触点,接入智能眼镜、录音卡片、耳机、键鼠、PC与平板等十余个品类、超30个品牌;应用层面推出"Buddy应用",由30多家伙伴共同打造覆盖20多个领域的品牌化AI工作台;开发者层面开放Skill、Expert、Connector,其中Connector兼容MCP与CLI两种思路。

不争应用争底座,WorkBuddy瞄准的是任务入口与上下文资产

速途网在现场全程跟完发布会后,最直观的感受是,WorkBuddy的野心并不在于开放本身,而是腾讯再次选择扮演连接者的角色。

过去二十年来,腾讯在社交、内容、支付领域不断验证着同一套打法——自建最底层基础设施,将上层空间留给合作伙伴。腾讯云副总裁、腾讯CodeBuddy与WorkBuddy负责人刘毅在发布会上也表示:"不是做一个更强的工具,而是做一个能承载所有工具的平台。"具体到开放架构上,则是三条路径齐头并进:硬件厂商"无需自建AI,也不必改变机型形态",行业伙伴"只需把自身情形、业务上下文、工具能力与品牌接进来"。

实际上,WorkBuddy作为底层平台,其核心资产主要体现在两方面:任务入口与上下文。所谓任务入口,是指当Agent代替用户分发任务时,谁接住了首条指令,谁便拥有了分配权。相比之下,上下文的价值更高,因为入口随时可能被替代,用户换个渠道发号施令即可;但上下文是用户在日常使用中逐渐积累的,会议纪要、项目进展、客户关系等均沉淀于此,一旦更换底座便意味着推倒重来,且使用时间越长、迁移难度越大。此次WorkBuddy与智能眼镜达成联名合作,实现了所见所闻内容的自动归档与可检索化,账号、任务、记忆及产出物在眼镜、手机、PC和Web端实现无缝同步。不难看出,谁掌控了这种跨端记忆,谁就扼住了用户数字工作流的咽喉。

这同样揭示了腾讯为何甘愿让生态的商业化进程放缓脚步。对于底层平台而言,优先跑通上下文的沉淀机制,远比尽早确立分账规则来得重要。反观阿里巴巴,其重心在于企业流程,试图依托钉钉的组织关系链来维系留存率;字节跳动则将筹码压在流量与协同效应上,凭借3.82亿月活的豆包来拉动新增。腾讯押注的则是跨端上下文,这一方向具备最大的想象空间,但也最难在短期内得到验证。

诚然,此种逻辑在科技行业内并非罕见。微软将Copilot嵌入Windows及Office体系,Salesforce则借助Agentforce锁定CRM数据,各方争夺的本质如出一辙:谁与业务上下文的距离最为贴近,谁就拥有最高的不可替代性。

面对三条生态发展路径,腾讯为何偏偏选择了最为艰难的那一条?

速途网认为,若不局限于单场发布会,而是将三家巨头近两个月来的举措横向对比,便会发现行业竞争的焦点已发生转移。如今的较量已不再是比拼谁的Agent执行任务更强,而是看谁能充当其他Agent的底层基石。

阿里巴巴走的是内嵌式路线。千问办公将QoderWork、MuleRun及悟空三者融为一体,直接扎根于钉钉2000万企业的组织关系及权限架构中,Agent无需另行开辟入口。该路径落地迅速,但代价是其生态边界完全受限于钉钉自身的边界。

字节跳动则选择了内循环策略。7月30日飞书并入豆包,8月24日TRAE与扣子加入,8月25日豆包工作面世,短短不到一个月内,便将所有生产力工具收归于同一品牌麾下,随后利用3.82亿月活的豆包进行导流。该路线的整合成本较低,但代价是仅能服务于自有体系内的资源。

腾讯选择的是第三条路线——外扩,这也是最具挑战性的一条。它将Agent Harness拆解为标准接口,以此换取硬件厂商的入口及行业伙伴的场景。其优势在于生态边界不受限于自有体系,劣势则在于每个环节均需逐一谈判,且任一环节都有可能被竞争对手以同等条件挖走。这同样解释了一个看似反常的现象:另外两家均采取先收拢后开放的策略,唯独腾讯在尚未完成收拢时便先行铺开。

更需引起注意的是,企业微信内部也孵化出了一款Agent。今年6月,企业微信的AI助理“大圆”开启内测,并于7月在WAIC上首次公开露面,其采用的同样是基于群聊、文档、会议及日程来自动解析用户诉求的路径;5.0版本之后,智能表格、智能文档及服务总结等功能也相继上线。换言之,腾讯实质上已掌握了一个根植于组织内部的Agent入口,未来它将如何与WorkBuddy进行分工,同样是一个亟待解决的难题。

此外,办公Agent构建生态的难度,在某个维度上甚至超越了移动互联网时代。当年的应用商店把持着分发渠道,用户安装何种应用由商店说了算;而Agent生态争夺的核心在于上下文,用户的会议记录、项目进度及客户资料零散分布于各家软件之中,各方皆意图聚合这些数据,却无人肯轻易交出。这也正是三家巨头率先采取的举措,均是死死攥住自有体系内资产的根本原因。

在这场发布会上,速途网察觉到的另一重信号并非腾讯吸纳了多少合作伙伴,而是办公Agent领域的角逐已由机型层面跃升至生态层面。未来半年内,关注的焦点不该是新增了多少品牌,而应聚焦于三个关键点:生态分账机制何时正式落地、硬件联名机型的实际出货规模,以及Buddy应用的增长趋势。

写在最后

此次发布会明确了清晰的战略抉择:腾讯致力于成为Agent时代的连接层,以开放姿态换取应用场景,凭借底座能力获取上下文信息。

然而,现在将WorkBuddy开放平台定义为“Agent操作平台”尚言之过早。在速途网眼中,它目前更类似于一条正处于管道铺设阶段的Agent总线,虽然接入的设备和行业应用数量已相当可观,但“水流”尚未真正输送至合作伙伴端。尽管WorkBuddy已通过出售积分实现变现,但合作伙伴的商业化路径仍处于搭建过程之中。总线架构本身具备内在价值,未来的核心看点在于合作伙伴能否借此获利、用户能否在该平台上采购到所需服务,这些因素将直接决定WorkBuddy与“Agent时代操作平台”之间的实际距离。返回搜狐,查看更多

本文来源:https://www.sohu.com/a/1072440078_114814
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Tencent WorkBuddy Open Platform Launches — How Close Are We to an “Agent Era Operating System”?

On September 2nd, Tencent WorkBuddy officially announced the launch of its open platform, including over 100 ecosystem partners at the first meeting, as well as the debut of 9 co-branded smart hardware, integration with 30+ industry applications, and the opening of three core capabilities to developers: Skill, Expert, and Connector.

At the press conference, Liu Yi, Vice President of Tencent Cloud and head of Tencent CodeBuddy and WorkBuddy, defined WorkBuddy's positioning very highly — to be "an operating platform for the Agent era, not to make a stronger tool, but to build a platform that can carry all tools."

Compared to Alibaba and ByteDance, Tencent is the earliest to take action, and the only one to simultaneously roll out hardware, applications, and developers. In this open ecosystem, WorkBuddy is actually betting on the underlying framework and contextual assets, and whether the opening will succeed depends entirely on whether partners can derive commercial benefits from it.

Within three months, the three giants have all entered the field. Has Tencent already pulled ahead of the pack?

The office Agent track in 2026 will only truly take shape in August.

On August 3rd, Alibaba announced the public beta of "Qianwen Office", which integrates QoderWork, MuleRun, and Wukong to cover over 20 million enterprises via DingTalk; on August 25th, ByteDance launched "DuoBao Work", and previously on July 30th, Feishu, and on August 24th, TRAE and Kouzi were all integrated into the DuoBao system.

Tencent laid the foundation earliest, with WorkBuddy appearing in March, nearly 5 months ahead of Alibaba and ByteDance. According to AICPB data, in July, WorkBuddy reached 11.1523 million MAU on PC clients, with a 304.4% increase month-on-month, ranking first in the AI office agent total list; ByteDance's TraeWork had 2.6599 million, Alibaba Qianwen Office had 607,900, and Baidu Dazi led the growth with 6.743 million MAU.

However, WorkBuddy's first-mover advantage does not seem to have turned into a generational gap. A test report by the Wall Street investment bank Jefferies showed that the comprehensive scores of Qianwen Office, DuoBao, and WorkBuddy were 95, 77, and 66 respectively. This indirectly indicates that WorkBuddy's lead relies more on entry point advantages and first-mover gains, rather than the quality of task execution itself.

Therefore, I saw Tencent WorkBuddy build an open ecosystem. In terms of hardware, around the five touchpoints of "listening, seeing, recording, chatting, and coordinating", it connects smart glasses, recording cards, headphones, mice, PCs, and tablets, covering over 30 brands; in terms of applications, it launched "Buddy Application", jointly created by 30+ partners to cover branded AI workbenches in 20+ fields; in terms of developers, it opens Skill, Expert, and Connector, where Connector is compatible with MCP and CLI.

Rather than competing for applications, WorkBuddy aims for the underlying framework and contextual assets.

After following the entire press conference on-site, my most direct impression is that WorkBuddy's ambition is not in opening per se, but in Tencent once again playing the role of a connector.

Over the past two decades, Tencent has repeatedly verified the same strategy in social, content, and payment — building the most fundamental infrastructure itself and leaving the upper layers to partners. Liu Yi also stated at the press conference: "Not to make a stronger tool, but to build a platform that can carry all tools." Specifically in the opening architecture, three paths are advancing simultaneously: hardware manufacturers "no need to build AI themselves, nor change the model form"; industry partners "only need to bring their own situation, business context, tool capabilities, and brand in".

In fact, WorkBuddy, as an underlying platform, has its core assets mainly in two areas: task entry and context. The so-called task entry means that when the Agent distributes tasks instead of the user, whoever catches the first instruction has the distribution right. The value of context is higher, because the entry can be replaced at any time — the user just issues the command via another channel — but the context is gradually accumulated during daily use, including meeting minutes, project progress, and customer relationships, all of which are settled here. Once the underlying framework is changed, it means starting over, and the longer the usage time, the more difficult the migration. This time, WorkBuddy’s collaboration with smart glasses has realized the automatic archiving and searchability of what is seen and heard, and accounts, tasks, memories, and outputs have been synchronized seamlessly between glasses, mobile phones, PCs, and web. Obviously, whoever controls this cross-device memory holds the throat of the user's digital workflow.

This also reveals why Tencent is willing to slow down the commercialization process of the ecosystem. For the underlying platform, prioritizing the establishment of the context accumulation mechanism is far more important than finalizing the revenue-sharing rules as soon as possible. Looking at Alibaba, its focus is on business process, trying to maintain retention through the organizational relationship chain of DingTalk; ByteDance bets on traffic and synergy, using 382 million MAU of DuoBao to drive new users. Tencent bets on cross-device context, which has the greatest imagination space, but is also the hardest to verify in the short term.

Of course, this logic is not uncommon in the tech industry. Microsoft embedded Copilot in Windows and Office, and Salesforce used Agentforce to lock in CRM data — the essence of the struggle is the same: who is closest to the business context has the highest irreplacability.

Faced with three paths for ecosystem development, why did Tencent choose the most difficult one?

I believe that if we compare the moves of the three giants over the past two months, rather than just looking at one press conference, it becomes clear that the focus of the industry competition has shifted. Now it’s not about who’s Agent executes tasks better, but who can serve as the underlying foundation for other Agents.

Alibaba is taking the embedded route. Qianwen Office integrates QoderWork, MuleRun, and Wukong directly into the organizational relations and permissions of 20 million DingTalk enterprises, so the Agent doesn’t need a separate entry. This lands quickly, but the cost is that the ecosystem boundaries are completely limited by DingTalk's own boundaries.

ByteDance is taking the internal circulation strategy. On July 30th, Feishu merged into DuoBao; on August 24th, TRAE and Kouzi joined; on August 25th, DuoBao Work came out — in less than a month, all productivity tools were brought under one brand and funneled via 382 million MAU. The integration cost is low, but the price is that it can only serve resources within its own system.

Tencent chose the third path — expansion — which is also the most challenging. It breaks down the Agent Harness into standard interfaces to exchange for entry from hardware manufacturers and scenarios from industry partners. The advantage is that the ecosystem boundaries are not limited to its own system; the disadvantage is that every link requires negotiation, and any link could be poached by competitors under better terms. This also explains a seemingly contradictory phenomenon: while the other two are adopting a “consolidate first, then open” strategy, Tencent is opening up before consolidation is complete.

More notably, Weixin Enterprise (WeCom) has also incubated an Agent. In June, WeCom AI Assistant “Dayuan” started beta testing and debuted at WAIC in July, using the same path of automatically parsing user intents from group chats, documents, meetings, and calendars. After version 5.0, smart tables, smart docs, and service summaries were added. In other words, Tencent already has an entry for Agents rooted in the organization — how it divides labor with WorkBuddy is itself a problem to solve.

Furthermore, building an ecosystem for office Agents is harder than the mobile internet era in one sense. Back then, app stores controlled distribution channels — users installed what the store said. Now, the core of Agent competition is context — meeting notes, project progress, customer data scattered across apps, each party trying to aggregate it, but no one wants to let go. This is why the first move of all three giants was to tightly hold onto their own system assets.

At this press conference, another signal I noticed was that the race for office Agents has moved from the model level to the ecosystem level. Over the next six months, the focus shouldn’t be on how many brands were added, but on three key questions: when will the ecosystem revenue-sharing mechanism land? What is the actual shipment scale of co-branded hardware? And how is the Buddy Application growing?

In conclusion: this press conference clarified a strategic choice — Tencent positions itself as the connective layer of the Agent era, trading openness for application scenarios, using the underlying framework to acquire context. However, defining WorkBuddy’s open platform as an “Agent operating platform” is premature — it currently resembles an Agent bus in the early stages of pipeline construction, with partners connected but not fully flowing. The real test lies in whether partners can profit from this, and whether users will buy services here — that’s what determines the distance between WorkBuddy and the “Agent era operating platform.”

This release clearly outlines a strategic decision: Tencent aims to become the connecting layer for the Agent era, trading openness for access to application scenarios while using its foundational capabilities to acquire contextual information.

However, it is premature to define the WorkBuddy Open Platform as an "Agent Operation Platform" just yet. In the eyes of the journalist, it currently resembles an Agent bus that is under construction, where devices and industrial applications are being connected, but the "water flow" is not yet reaching the partners. While WorkBuddy has monetized through point sales, the partners' commercialization paths are still in the design phase. The bus architecture holds intrinsic value, and its future success will depend on whether partners can profit from it and whether users can procure services through this channel — these factors will determine the true distance between WorkBuddy and the title of "Agent Era Operation Platform."

This article comes from: https://www.sohu.com/a/1072440078_114814. If there are any copyright issues, please contact us at 13951431913#139.com or submit a comment on the right.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry