(Yicai) Aug. 25 -- ByteDance launched its standalone artificial intelligence office product Doubao Work today, as China’s internet giants pivot toward workplace AI after spending billions of yuan to win consumer users.

Doubao Work, an agent-based product deeply integrated with ByteDance’s workplace collaboration platform Lark, can autonomously break down tasks, call tools, and execute complex workflows. With user authorization, it can even operate computers across different software applications. Yicai also learned that the teams behind ByteDance’s AI coding product TRAE and AI agent development platform Coze will be fully integrated into the Doubao ecosystem.

Major internet companies are consolidating their AI product lines around workplace applications, seeking clearer demand and more sustainable commercialization after costly consumer campaigns delivered short-lived traffic gains. But the outlook remains uncertain, as Chinese companies are less accustomed to paying for Software-as-a-Service products and enterprise AI must prove it can cut costs and boost efficiency.

The shift is already reshaping product lineups across the industry. Tencent has moved QClaw, an AI agent that can operate computers remotely, into the same structure as workplace agent WorkBuddy. Alibaba has combined QoderWork, Wukong, and MuleRun into Qwen Office, bringing various AI workplace tools under one platform.

From Consumer Traffic to Office AI

Around the Chinese New Year holiday in February, these internet giants were still spending heavily on consumer AI products.

Tencent’s general-purpose AI assistant Yuanbao launched a CNY1 billion (USD148.8 million) cash giveaway campaign, while ByteDance’s chatbot Doubao secured an exclusive partnership with China Central Television’s Spring Festival Gala. Alibaba’s AI assistant Qwen rolled out its CNY3 billion “Spring Festival Treat Campaign,” offering users free food, shopping, and entertainment.

The massive spending quickly translated into traffic peaks. Doubao’s daily active users reached 145 million, while Qwen’s hit 73.5 million and Yuanbao’s 40.5 million. But traffic faded almost as quickly as it arrived. After the marketing campaigns ended, their rankings across major app stores rapidly declined, while user retention and actual conversion fell far short of expectations.

The underlying issue is that the commercial logic of the AI era has fundamentally changed. In the mobile internet era, burning money to acquire traffic and monetizing it through advertising worked because marginal costs approached zero. For large AI models, however, every user interaction consumes substantial computing resources. More users therefore mean higher costs, without a comparable network effect to spread them.

The financial toll of that AI push is already showing. Tencent’s non-International Financial Reporting Standards operating profit reached CNY75.6 billion (USD11.3 billion) in the first quarter of this year, up 9 percent from a year earlier, according to the company’s financial results.

Excluding revenue, costs, and expenses associated with five new AI products -- Hy, Yuanbao, CodeBuddy, WorkBuddy, and QClaw -- non-IFRS operating profit was CNY84.4 billion, up 17 percent year on year. The CNY8.8 billion gap indicates the impact of the five new AI products on operating profit.

Such costs are pushing technology giants toward applications where customers are more willing to pay.

Office AI Faces Commercialization Challenges

But workplace AI brings its own commercialization challenges.

Zhang Fa’en, chief scientist at enterprise AI solutions provider AInnovation Technology Group, previously told Yicai that the US SaaS ecosystem is highly mature and agents essentially replace “humans operating software,” while Chinese companies have yet to fully develop the habit of paying for SaaS products.

Enterprise procurement is ultimately about cutting costs and improving efficiency rather than experimenting with new technologies, Zhang said, adding that AI must be deeply integrated into specific business processes to create real value, placing high demands on internet giants’ industry expertise, information security, and customization capabilities.

AI office products therefore still need to prove they can deliver tangible benefits to enterprises, Zhang concluded.

Editor: Emmi Laine