--- title: "Tencent, BABA, ByteDance: Who will call the shots in AI productivity?" type: "Topics" locale: "en" url: "https://longbridge.com/en/dolphin/post/44318101.md" description: "Chat boxes don’t monetize well. Agents are taking the baton." datetime: "2026-10-14T03:29:29.000Z" locales: - [en](https://longbridge.com/en/dolphin/post/44318101.md) - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/44318101.md) - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/44318101.md) author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)" generator: "portal-rs" --- # Tencent, BABA, ByteDance: Who will call the shots in AI productivity? In the prior piece (see [link](https://longbridge.com/en/dolphin/post/44306113)), Dolphin Research broke down the shift in AI entry points: chatbots deliver 'advice' rather than 'outcomes', making ROI opaque for users and limiting pay conversion. For agents, the moat is neither the model nor the harness, but user usage data and owned ecosystems. So why are consumer-facing agents for billions of users stuck behind three walls: ecosystem, privacy, and cost? How large is the enterprise office market? Among Tencent Workbuddy, Alibaba Qwen Office, and ByteDance Doubao Work, who has the edge? More below. Details follow: **I. Personal Agents vs. Office Agents: Long-term ideal vs. near-term reality** By analogy with the internet, consumer internet value dwarfs enterprise internet. If a consumer agent becomes the primary gateway for e-com and OTA, the ad and transaction-take rate upside is huge — in 2026, each US DAU could generate roughly $300 in ad revenue. **But there are three walls today:** **(1) Ecosystem wall:** When an agent invokes services on third-party platforms, it can bypass splash ads and recommendation feeds, causing those platforms to forgo ad and commission revenue, so AI-native gateways tend to face resistance from incumbent platforms. This is more acute in China. A case in point: upon Doubao Phone’s launch, it faced a coordinated block by WeChat, Taobao, Alipay, and multiple banking apps, while Meta Muse encountered deep-link bans from Amazon in N. America after its Sep launch. **(2) Privacy wall:** Personal agents need bank logins, chat history, and payment permissions. After Muse launched, incidents followed around 'unauthorized message reads' and 'human-assisted calls without adequate disclosure', while its promised 'encrypted VM' won’t roll out until year-end (a standard isolation VM is live). Open-source OpenClaw exposed 135k public instances and thousands of malicious skills, prompting Gartner to advise enterprise bans. Without deep integration into daily life, personal agents struggle to deliver truly effective AI services. This remains a fundamental adoption barrier. **(3) Cost wall:** Mass-market stickiness for personal agents (time spent and frequency) remains modest. Even for Muse, still riding marketing momentum one month post-launch, DAU/MAU only matches ChatGPT despite social traffic support, far from a true assistant benchmark (comparable to social platforms), making near-term monetization challenging. Muse isn’t cheap: each user gets a dedicated VM, and the platform bears token costs for free and low-tier paid users. Per Morgan Stanley, assuming 10–20 tasks/day per user (including always-on agents), **monthly service cost per user is ~$15–37 (CPU sandbox + GPU compute).** **Another expert channel similarly estimates ~$30–50 per user per month.** Using $30/month as a reference implies $360 per user per year. Yet the current ad ARPU at Meta’s two major platforms, Facebook and Instagram, is still below $360. Moreover, the token usage for those 20 tasks sits within the free tier, so users need not upgrade to a paid Muse plan. ![IMG_262](https://pub.pbkrs.com/cms/2026/0/nRPonKQ9ZXmSw4TfBHNar4347byHBFf8.jpg?x-oss-process=style/lg) By contrast, for enterprise office scenarios, the long-term TAM has clearer anchors in efficiency gains and displacement of traditional software. **At its 1H26 results, Zhipu projected a $5tn global TAM for co-working (vs. $100bn for chat and $500bn for coding), multiples larger.** **Another approach is to benchmark Office directly,** using Office’s paid penetration (46% = 460mn seats out of 1bn knowledge workers) and $40 per seat per month, yielding a **total addressable market of $216bn.** Dolphin Research believes both projections are off. $5tn would equal roughly one-fifth of total wages for the 1bn+ global white-collar workforce and 3–4x the SaaS market, implying a seismic shift in how production is organized. The $200bn+ estimate undercounts the Office base (ignores pirate demand) and understates the value AI agents can capture (beyond text processing). Dolphin Research estimates TAM from two angles: labor cost savings/productivity gains and software displacement, for reference. **(1) Individual productivity & labor cost reduction** First, value via labor substitution: estimate the share of work hours an agent can assume, then compute theoretically saved wages. Given offline coordination costs, transition frictions, and ESG-related pressures, firms may expand output rather than only pursue layoffs, so we introduce a revenue-share factor to proxy willingness to pay. **Final TAM = total wages of knowledge workers × share of hours assumable by agents × vendor revenue share.** Assumptions by market and resultant estimates are shown below, summing to a $406bn TAM across regions. **(2) Enterprise software displacement** The other TAM source is displacement of existing enterprise software, or at least a reduced wallet share within IT budgets. Gartner suggests that by 2030, agents could cannibalize about 10% of traditional SaaS spend (est. $1.44tn in 2026; at 12% CAGR to $2.3tn by 2030), or roughly $230bn. Dolphin Research views this as conservative. Based on migration difficulty by software category, we estimate a higher aggregate displacement of **16%: a $227bn TAM on 2026 spend and $356bn by 2030** (using Gartner’s growth for total software spend). Office collaboration and data analytics/BI are easiest to migrate, assumed at 40%–50%. Infrastructure tools have the lowest displacement at 10%, not due to difficulty but because agents can leverage them as-is. ![IMG_264](https://pub.pbkrs.com/cms/2026/0/tzobTaLiPgicbsvqZZLHu1bNgXTGNw1Z.jpg?x-oss-process=style/lg) **(3) Global TAM after aggregation** On a 2026 base, labor savings plus software displacement yields $406bn + $226.8bn = $632.8bn. This sits between the $216bn Office-based benchmark and the $5tn optimistic view by Zhipu. Some double-counting may exist — a firm’s reduced software bill could overlap with labor savings. **Even so, the market should still approach ~$500bn.** Applying the same method to China, labor savings of $33bn plus $11.5bn yields a $44.5bn TAM, or roughly RMB 320bn. ![IMG_265](https://pub.pbkrs.com/cms/2026/0/aRow9TNvQmCwe98MNb99rYLkDYJfsFGq.jpg?x-oss-process=style/lg) **II. AI office: who is best positioned to be the foreman?** Focusing on China, we look at the core players: Tencent Workbuddy, ByteDance Doubao Work, Alibaba Qwen Office, and model vendor Zhipu Zcode. As a high-potential app layer to monetize the hyperscalers’ massive compute bets, **Dolphin Research expects fiercer and more sustained competition vs. the early-year chatbot land grab.** But the giants are likely moving away from pure subsidy-led acquisition. **Instead, they will lean on existing ecosystems and channels, and compete on product velocity to reach users fast, drive retention, and accumulate core enterprise and user data assets** — building a flywheel where scale begets strength. It is still early to call an endgame winner, given the AI office market only went all-in mid-year. That said, using the framework from the prior piece, we can assess each player’s odds today. The four-layer stack below sketches the key operating layers. **On completeness, Tencent, ByteDance, and Alibaba all have full-stack ecosystems, while Zhipu lacks a 100mn+ entry product for top-of-funnel traffic, a disadvantage.** Among AI-native office agent platforms, **Workbuddy currently leads on user scale,** aided by Tencent’s consumer traffic and the HY-series’ compelling free-token economics. ![IMG_267](https://pub.pbkrs.com/cms/2026/0/FvAYDmBXf27HnadPCJLgpT2b8NaPimpN.jpg?x-oss-process=style/lg) How might the race evolve? We compare using the key factors summarized previously: **1\. Context advantage** Pure tech is not a decisive moat. A competent, doer-level LLM is the ticket to play — it need not top leaderboards, but must complete long tasks reliably at low unit cost. In China, where models are cheaper and catch-up is faster, tech alone is an even weaker moat. User usage data is crucial. It improves post-training, hardens playbooks and memory, and deepens understanding of enterprise workflows, which upgrades UX and raises switching costs. To build context, either leverage high-frequency traffic within your ecosystem to accumulate usage traces, or bring deep industry know-how and integrations. Both routes compound over time. ![IMG_268](https://pub.pbkrs.com/cms/2026/0/zk6EP5u9KLDHMK3YHqFJJ9v1kmXyiizF.jpg?x-oss-process=style/lg) **On context, Dolphin Research ranks: Alibaba > ByteDance >= Tencent.** **(1) Alibaba best-in-class, with entry traffic plus industry know-how.** DingTalk serves 26mn+ orgs nationwide, with the deepest penetration across traditional industries, government, and education among the three. Their approval flows, org structures, and time/attendance live natively on DingTalk — the most natural 'enterprise context'. Qwen Office productizes that context into 25 enterprise modules and 20+ connectors, plus the MuleRun expert network and a skill marketplace. Limitations: DingTalk skews to SMEs and is industry-fragmented, capping depth per vertical, and its know-how is concentrated in general workflows (approvals, attendance, reimbursements), with less depth in specialized domains vs. Feishu’s top-tier accounts. **(2) ByteDance in the middle, with higher-quality customers** Feishu’s base clusters in internet, NEV, new consumption, and advanced manufacturing — firms with 'advanced management'. These clients migrate core processes to Feishu (e.g., NEV R&D collaboration, tea chain store ops), forming high-value, replicable best practices. ByteDance invests most in on-site delivery. Our checks indicate an FDE team of 60–70 people, improving internal efficiency and crafting market-facing acquisition solutions. Doubao Work has rolled out vertical playbooks across autonomous driving, AI model companies, AI education, AI pharma, and AI for Science. Constraints: limited entry traffic, as Feishu MAU is ~1/7 of DingTalk, and narrower industry coverage. Deep tying to Feishu hampers cross-ecosystem delivery, shrinking multi-platform options for enterprises. **(3) Tencent currently leans on an open ecosystem, while building a data asset via traffic** Despite Workbuddy’s fastest pace among AI-native office platforms — with DAU reportedly above 6mn in Sep — Tencent lacks both comparable entry traffic it can leverage and deep industry knowledge. WeChat Work focuses more on external comms and CRM, with shallow embedding in internal processes; it is often viewed as the 'weakest enterprise modules, most neutral ecosystem' among the three. Tencent’s approach is thus more open than Alibaba and ByteDance: 'we provide agent, model, and distribution; partners bring industry data and know-how'. The familiar open-ecosystem playbook is back. a) On entry supplementation, Workbuddy integrates with 30+ hardware brands across 10+ categories, including smart glasses, recorders, mics, and headsets. b) On industry know-how, the Workbuddy open platform has onboarded 100+ partners across 20+ industries, with 30+ in deep collaborations. ![IMG_269](https://pub.pbkrs.com/cms/2026/0/D8qTMk8ARhC5ijEfLvCGgm8yidYEWMkL.jpg?x-oss-process=style/lg) Workbuddy’s edge today lies in external-facing scenarios like financial marketing and private-domain retail, where WeChat Work directly reaches WeChat users. The know-how and data are unique versus the other two. One caveat: non-exclusive partnerships mean these partners can also connect to DingTalk and Feishu. This dilutes differentiation. **That said, Dolphin Research sees Workbuddy’s longer-term opportunity as more pronounced: sustained accumulation of consumer-side usage data, and a unique Tencent-ecosystem advantage — integrating personal-agent and work-agent data — aided by the rise of OPC (one-person companies) in the AI era.** **2\. Business model differences** Ecosystem advantages and weaknesses shape monetization paths. Penetration is still early; everyone keeps subscriptions, but no one in China expects office agents alone to pay the bills — bundling with ecosystems and legacy products is key. **For smoother monetization via current models, Dolphin Research ranks: ByteDance > Alibaba > Tencent.** Over time, as data assets and context compound, China’s AI office market — potentially RMB 300bn+ — could still become a low-friction lubricant for Big Tech’s AI-era transition. **(1) Alibaba: monetization via 'Cloud + DingTalk + MaaS'** Alibaba starts with enterprise procurement, bundling cloud with DingTalk to win top-down, aiming to boost DingTalk retention, then monetizes through seat sales plus backend token and Alibaba Cloud consumption. **(2) Tencent: monetization via 'ecosystem take rate + cloud'** With strong consumer traffic, Workbuddy sources not only from WeChat Work but also broader WeChat channels and off-platform ad buys targeting white-collar users. Given the large share of 'retail' users, it leans more on subscriptions and marketplace take rates. **(3) ByteDance: monetization via 'enterprise IM + cloud'** ByteDance takes a middle path: stitching AI into Feishu to fortify the entry, while also running C-end acquisition. Monetization bundles with Feishu and Volcano Engine — Feishu for enterprise context and seat revenue; Volcano for compute and token revenue; Doubao Work lifts ARPU and stickiness in between. The org mirrors this: Feishu product moved under Doubao, Feishu sales under Volcano. **Consistent with Volcano’s sales style, Doubao Work runs aggressive motions — e.g., free three-week, 100+ seat trials for enterprises buying Volcano compute, then converting to paid.** **Conclusion: AI office is the key tiller for internet incumbents’ turn** The AI office market is large (Dolphin Research sees long-term global TAM of $500bn+ and China at $45bn). Thanks to expanding AI capabilities, it may monetize tokens faster than consumer assistants. **Incumbent internet giants, under near-term AI ROI pressure and long-term transformation needs, will not cede this battleground.** For now, Workbuddy enjoys a first-mover advantage. But on core vectors (context, ecosystem, monetization), **it still needs to shore up model performance and industry know-how to slow fast followers.** Qwen Office and Doubao Work have credible catch-up — even leapfrog — potential, not only on user counts but via deep enterprise integration. **Whoever amasses more workflow experience and related data assets will build the most durable moat.** **Workbuddy’s unique edge lies in bridging personal-life data and work data,** a longer-term moat given complex data isolation and permissioning across software and hardware. This will take time to harden. \ [Risk Disclosure and Disclaimer: Dolphin Research](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [META.US](https://longbridge.com/en/quote/META.US.md) - [02513.HK](https://longbridge.com/en/quote/02513.HK.md) - [002230.CN](https://longbridge.com/en/quote/002230.CN.md) - [00001.HK](https://longbridge.com/en/quote/00001.HK.md) - [CKHUY.US](https://longbridge.com/en/quote/CKHUY.US.md) - [HZAD.SG](https://longbridge.com/en/quote/HZAD.SG.md) - [00700.HK](https://longbridge.com/en/quote/00700.HK.md) - [09988.HK](https://longbridge.com/en/quote/09988.HK.md) - [BYTED.NA](https://longbridge.com/en/quote/BYTED.NA.md) - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) ## Comments (1) - **NewUser_mCQEjBu · 2026-10-14T23:08:30.000Z**: Yes! --- > **Disclaimer: This article is for reference only and does not constitute any investment advice.**