NextFin News — Four major Chinese technology companies have moved in quick succession to turn large-scale AI assistants into tools that can be charged for. Between June and late August 2026, ByteDance, Alibaba, Tencent and Baidu each advanced dedicated office or productivity agents, confronting the same underlying problem: two years of free access trained users to treat AI as a zero-cost utility, while every inference continues to incur real compute expense.

ByteDance’s Doubao remains the largest consumer AI application by active users. Third-party trackers placed its monthly active users above 380 million by mid-year. On June 24 the product introduced a three-tier professional subscription starting at 68 yuan per month, with higher tiers at 200 and 500 yuan, unlocking greater quotas and the 2.1 Pro model for office-style tasks.

Free access continues for basic use. Later in August the company released Doubao Work, an agent-oriented product with a desktop client and deep Feishu integration that can draft documents, tables and presentations while drawing on authorized enterprise context. Organizational changes accompanied the push: Feishu product teams were folded into Doubao, followed by further consolidation of related coding and agent platforms under the same umbrella.

Alibaba took a different route. Qwen’s consumer growth was already steep; the application then gained further traction by linking directly into Taobao, Alipay and other ecosystem services, turning conversational assistance into completed transactions rather than pure advice. On August 3 the company opened public beta of QwenWork (千问办公), an integrated agent platform formed from earlier internal tools and powered by the newly released Qwen3.8 series.

The product supports desktop, cloud and collaboration agents and has since connected to DingTalk, Feishu and WeCom. Monetization leans on usage credits and enterprise subscriptions rather than persuading individual users to pay solely for chat capacity.

Tencent presents two contrasting experiments. Yuanbao, heavily promoted during the Lunar New Year period, saw rapid early adoption that later cooled according to third-party measurements of monthly actives and session length. In parallel, WorkBuddy, a desktop-focused agent launched in public beta in March, has recorded stronger retention metrics.

Analyst and industry data cited monthly active users in the low tens of millions and daily actives well above 10 million, with high DAU-to-MAU ratios that suggest habitual workplace use. Features emphasized local file handling, structured deliverables and controlled permissions—attributes useful for government and enterprise environments. Management commentary has indicated the product remains in an investment phase even after some enterprise pricing adjustments.

Baidu has pursued a multi-product matrix. After consolidating office AI efforts, it fields Baidu Mate (百度搭子) as a general-purpose agent, Kuku AI (库库AI, formerly GenFlow) oriented toward cross-device automation, and Miaoda for no-code application generation. Desktop rankings in mid-2026 placed Baidu Mate among the leading office agents by monthly actives and growth rate.

Company disclosures continue to show that a substantial share of AI-related revenue still derives from cloud infrastructure and related services rather than pure application subscriptions, while the firm highlights full-stack capabilities spanning chips, frameworks and models for institutional customers.

The competitive timeline is compressed. WorkBuddy’s early public beta, Doubao’s professional tier in June, QwenWork’s August launch, Kuku AI’s formal branding and Doubao Work’s late-August release collectively mark the transition from pure consumer chat scale to workplace embedding.

Office tasks offer clearer economic logic than open-ended conversation: a completed document, spreadsheet or presentation is a verifiable deliverable that can support pricing. Success is increasingly measured by task completion rates rather than raw model size or peak monthly actives.

Challenges remain common across the field. Converting consumer chat habits into reliable workflow dependence requires permissions, security, procurement processes and consistent multi-step execution. Continuous generation raises unit costs that free-era scale only magnified. Consistency across long agent sessions and automated content safety further constrain rapid commercialization.

The companies that solve embedding and measurable completion most effectively will hold the stronger claim on pricing power as the free-usage period recedes.