虎嗅

ByteDance is becoming more and more well-behaved (or “mannered”)…

原文:字节越来越善变了

Summary of Key Points

To fully commit to its AI strategy, ByteDance has made significant reorganizations within its B2B business: the Lark product team has been merged with DouBao (under Zhao Qi's leadership), and Lark’s marketing, sales, and customer service functions have been integrated into HuoShan Engine (coordinated by Tan Dai). The underlying rationale for these changes is that AI has become the “soul” of office tools, replacing traditional collaborative capabilities. As a result, Lark has been downgraded from an independent business unit to a specialized tool within DouBao’s larger model framework. This move aims to address issues such as internal resource dispersion and common challenges associated with large companies, while also competing with Tencent and Alibaba in the AI-powered office space.

Detailed Explanation

1. AI as the New Soul of Office Tools: Lark Moves from “Protagonist” to “Supporting Role”

In the past, companies purchased office software for its collaborative features—tools like chat, cloud documentation, and video conferencing—that made Lark a favorite among emerging internet companies. However, the AI era has completely changed the landscape. Today, businesses are seeking “intelligent productivity,” which involves tasks being automatically broken down and executed by AI (using large models). Lark’s AI features, such as smart meetings and task management tools, rely entirely on DouBao’s large model; Lark itself no longer possesses these capabilities independently. With ByteDance’s decision to create a unified AI-powered productivity platform, Lark’s status as a standalone product has been diminished, transforming it into a plugin for DouBao’s core models.

Tencent and Alibaba are also taking similar steps: Tencent has integrated QClaw into its cloud services, while Alibaba has consolidated three of its office products under the TongYiQianWen large model, making the large models the centerpiece and office tools the secondary components.

2. DouBao Takes the Lead: Building a Strong Foundation for the C-Segment and Leveraging Lark in the B-Segment

DouBao’s success is based on two key factors:

  • A Large C-Segment Base: As of March 2026, DouBao had 345 million monthly active users (the highest among domestic AI apps), with an average daily token usage of 180 trillion (a 1500-fold increase from 2024), processing 2.08 billion lexical requests per second—a significant source of traffic.
  • Clear Commercialization Path: DouBao initially attracted C-Segment users with free services before launching premium subscription options (68, 200, and 500 yuan per month). For the B-Segment, it integrates with Lark’s ecosystem to convert Lark’s corporate clients into DouBao users, offering features such as enterprise knowledge libraries and data security. By merging with Lark, DouBao can seamlessly integrate its large model capabilities into office processes, creating a closed-loop of productivity that spans both the C- and B-Segments.

3. Lark’s Sales Integration into HuoShan Engine: Reducing Internal Frictions and Providing Comprehensive Solutions

Previously, Lark and HuoShan Engine had separate sales teams targeting the same corporate clients, leading to redundant communications and resource competition, resulting in high customer acquisition costs. The integration now allows for a unified approach that offers customers “cloud computing power + large models + office software” as a single solution. This not only streamlines sales but also reduces internal inefficiencies within ByteDance’s three-tier business architecture:

  • HuoShan Engine: The foundation (computing power, cloud services);
  • DouBao: The core (model capabilities);
  • Lark: The application layer (implementation of office solutions).

4. ByteDance’s Efforts to Streamline and Focus on AI

With over 100,000 employees, ByteDance has faced typical issues associated with large companies, such as rigid departmental structures, slow decision-making, and resource waste. The reorganization aims to overcome these challenges:

  • The product team has been merged into DouBao to eliminate duplicate AI development efforts.
  • The sales teams have been integrated into HuoShan Engine to eliminate internal competition.
  • In terms of management, Lark’s leader Xie Xin now reports directly to Zhao Qi, signaling that performance is more important than seniority within the context of the AI strategy.

Liang Rubo previously mentioned the need to “escape the gravity of mediocrity,” and this reorganization reflects the company’s focus on concentrating resources on AI as its core competitive area.

5. HuoShan Engine’s Strengths and Challenges

HuoShan Engine leads the domestic MaaS market in terms of API usage (49.2% according to IDC reports), but it faces two major drawbacks:

  • Internal Traffic Impact: Its API usage includes traffic from internal products like TikTok and DouBao, reducing the “value” of its metrics.
  • Low Commercial Conversion: In 2025, HuoShan Engine’s revenue was only about 20 billion yuan, half that of Tencent Cloud and one-fifth of Alibaba Cloud’s. This is due to its low pricing strategy (thousands of tokens per use), which may not be sustainable in the long term. Additionally, its infrastructure (IaaS) is inferior to that of Tencent Cloud and Alibaba Cloud in terms of data center layout and computing power allocation.

By integrating Lark, ByteDance hopes to leverage the essential need for office software among businesses to drive sales of HuoShan Engine’s cloud services and models, thereby compensating for its weaknesses.

Conclusion

ByteDance’s adjustments essentially align all B2B resources towards AI: DouBao serves as the core, Lark provides the necessary applications, and HuoShan Engine forms the foundational technology. This move is both a response to industry competition (with Tencent and Alibaba also focusing on AI-powered office solutions) and a necessary step to overcome internal challenges. However, HuoShan Engine’s limitations in commercial conversion and infrastructure remain significant hurdles on its path to success in the AI for Business (AI ToB) market.