ByteDance has placed one of its most trusted executives, Zhao Qi, at the helm of a major organizational restructuring that merges its Feishu workplace software team into its Doubao AI division. This marks ByteDance's largest reorganization of its business-to-business operations since 2021, and it signals the company's determination to compete aggressively in the emerging AI office software market alongside rivals Alibaba and Tencent.

At ByteDance's all-hands meeting, Zhao Qi, the overall head of the Doubao product line, officially stepped into the spotlight as the architect of this strategic shift. The move consolidates ByteDance's scattered AI product lines and focuses them on enterprise customers, a departure from the consumer-focused growth strategies that have defined Zhao Qi's career.

Who Is Zhao Qi and Why Does His Track Record Matter?

Zhao Qi is not a typical AI researcher or engineer. He holds a doctorate in computer science from Peking University but built his career as a product manager focused on growth and monetization. His resume reads like a masterclass in scaling consumer platforms. He worked at Douban, the Chinese social network for literature enthusiasts, then moved to Wandoujia, an app distribution platform, before joining Chelaile, a real-time public transit app that now serves 334 million registered users across 488 Chinese cities.

When Zhao Qi joined ByteDance in 2017, he took charge of the company's "middle platform," a centralized system that provides shared capabilities to all of ByteDance's business divisions. Think of it as a shared engine that powers multiple products without each one needing to build from scratch. Under his leadership, Douyin, ByteDance's short-video app, exploded from roughly 30 million daily active users in late 2017 to 600 million by 2020.

Later, when ByteDance's advertising platform Pangle faced a crisis in 2021 after the Chinese government shut down the K12 education sector, Zhao Qi stabilized the business by shifting from simple "traffic selling" to precision data analysis. He calculated every ad slot, display, and click to maximize revenue per thousand impressions, expanding Pangle's scale significantly despite losing its largest advertiser category.

"The large language model boasts huge market potential, so he does not worry about economic returns, but revenue still remains a critical metric," Zhao Qi stated at ByteDance's all-hands meeting.

Zhao Qi, Head of Doubao Product Line at ByteDance

What Makes This Organizational Shift So Significant?

The integration of Feishu into Doubao represents a fundamental strategic choice. Feishu, ByteDance's enterprise collaboration platform, now reports directly to Zhao Qi instead of operating as an independent division. This consolidation puts ByteDance in direct competition with Alibaba's Qwen Office and Tencent's WorkBuddy, both of which are backed by their respective companies' massive enterprise customer bases.

The timing is crucial. All three Chinese tech giants made nearly identical strategic decisions in the same summer: integrate their AI product lines and focus on the office scenario. This is not a casual market move; it reflects a recognition that AI office software represents the next major battleground for enterprise revenue.

However, Zhao Qi faces an unprecedented challenge. Every achievement in his career has come in the consumer market, where success is measured by daily active users and growth velocity. Enterprise software operates under completely different rules. In the B2B world, the metrics that matter are customer success, contract renewal rates, and customized services, not user acquisition speed.

How Is ByteDance Approaching AI Model Development Differently?

While ByteDance reorganizes its business structure, it is also making bold technical bets. CEO Liang Rubo acknowledged at the company's all-hands meeting that the gap between ByteDance's Seed large language model and leading overseas models is widening. Rather than taking shortcuts, ByteDance is pursuing an expensive, high-risk strategy.

ByteDance founder Zhang Yiming explicitly opposed using a technique called "distillation," where companies feed queries to powerful models like Claude and use the outputs to train cheaper systems. This policy dates back to 2023, years before US sanctions threats made avoiding distillation politically useful. ByteDance is the only major Chinese AI lab that Anthropic has not accused of distillation, a distinction that reflects either unusual foresight or genuine research principles.

Instead of borrowing capabilities from competitors, ByteDance is discussing training a model with over 5 trillion parameters, roughly double the size of Alibaba's Qwen and Moonshot AI's Kimi K3. This approach requires massive computing resources, sophisticated data engineering, and organizational coordination, but it aims to create genuinely novel capabilities rather than copying existing ones.

• Distillation Rejection: ByteDance banned the practice of using competitor models as training data in 2023, a policy that predates current US regulatory pressure and reflects a commitment to independent development.
• Parameter Scale Ambition: The company is planning a model exceeding 5 trillion parameters, roughly twice the size of competing Chinese models, to achieve breakthrough capabilities rather than incremental improvements.
• Technical Leadership: The project will be led by Xiang Liang, head of Seed Foundation, and Shen Ke, head of large language model pre-training data, both veterans of ByteDance's search and advertising systems.
• Revenue Pressure: ByteDance's AI revenue is heavily skewed toward video generation, which depends on industry cycles; language models, especially coding capabilities, offer higher-value enterprise customers with greater stickiness.

Why Does ByteDance's Coding Gap Matter So Much?

One of the most painful gaps in ByteDance's AI portfolio is coding capability. During the Chinese New Year period, Anthropic's Claude quickly captured the market of programmers and enterprise customers with superior coding performance. Zhipu and Moonshot AI, with continuously improving coding abilities, saw their annual recurring revenue exceed 1 billion and 300 million US dollars respectively. ByteDance, by contrast, missed this critical window.

This matters because coding and programming represent the highest-value use cases for enterprise AI. Developers and companies building software are willing to pay premium prices for models that genuinely improve productivity. Video generation, by contrast, depends on the health of short-drama and e-commerce industries, which follow unpredictable cycles.

ByteDance's token consumption for Doubao reached 180 trillion in June, below the internal target of 250 to 300 trillion. More than half of that consumption comes from video generation models like Seedance and Seedream, while language models represent a smaller proportion. As the short-drama industry matures, token consumption growth is slowing, forcing ByteDance to build stronger language model capabilities to hit its 400 billion yuan revenue target for 2026.

Can Zhao Qi Replicate His Consumer Success in Enterprise Markets?

The fundamental question is whether Zhao Qi can transfer his proven playbook from consumer products to enterprise software. His entire career has been built on understanding how to acquire users at scale and monetize through advertising or traffic sales. Enterprise software requires different skills: building long-term customer relationships, ensuring successful implementations, and maintaining high renewal rates.

Zhao Qi has never suffered a defeat in his career at ByteDance, according to company records. But the AI office market is crowded with well-funded competitors backed by massive enterprise customer bases. Tencent's WorkBuddy has the personal supervision of Pony Ma, Tencent's founder. Alibaba's Qwen Office is integrated with DingTalk, which already serves millions of enterprise customers.

ByteDance's Doubao has reached 382 million monthly active users, ranking second globally behind ChatGPT and well ahead of competitors like Qwen at 166 million and DeepSeek at 127 million. This consumer strength provides a foundation, but translating consumer adoption into enterprise revenue requires a fundamentally different approach.

The stakes are extraordinarily high. ByteDance is betting that Zhao Qi can succeed in a market where he has no proven track record, while simultaneously refusing to take shortcuts through model distillation and instead investing in expensive, uncertain large-scale model training. Whether this strategy represents visionary foresight or an expensive mistake will become clear only when results arrive.