Summary of Key Points
Recently, both ByteDance and Alibaba have made significant adjustments to their office products (Lark and DingTalk): ByteDance split Lark into two parts, with one half integrated into the AI model team (DouBao) and the other into the cloud team (HuoShan Engine); Alibaba combined three of its internal AI assistants into “QianWen Office,” which is now interconnected with DingTalk. The reason behind these changes is the emergence of AI Agents, which have completely rewritten the rules of competition in office software. In the past, the focus was on having more standardized processes and a wider range of features; now, it’s about whose AI can more accurately help companies solve personalized problems. Although the approaches differ (ByteDance is consolidating resources, while Alibaba is dispersing them), both face the same challenge: how to transform the user data and organizational relationships accumulated over the past decade into advantages in the AI era, while also finding new ways to generate revenue.
1. Old Rules Are Obsolete: AI Agents Change the Game for Office Software
The competition between Lark and DingTalk used to be like that between two companies selling “success templates”: DingTalk moved offline management tasks (such as time tracking and expense reporting) online, while Lark turned ByteDance’s own work methods (like document collaboration) into products, essentially suggesting that companies follow someone else’s successful processes. The company with more templates, users, and large clients would win.
However, with the advent of AI Agents, the game has changed. Companies no longer need to follow pre-set templates; they can simply tell the AI how their operations work and what problems need to be solved, and the AI will create a workflow tailored to their needs. For example, if you ask the AI to organize last week’s meeting minutes, synchronize them with the project team, and generate a to-do list, it can do all this directly across different applications without you having to manually open documents, send messages, or create tables.
The evaluation criteria have also shifted from “number of features and users” to “task completion rate”: Can the AI understand your needs? Can it work across different software? Can it become more intelligent with use? This transformation has turned office software from a “toolset” into an “intelligent assistant,” and past advantages (such as a wide range of features) may now be disadvantages, as they can become more complex to use.
2. Legacy Assets: A Moat or a Burden?
After ten years of development, Lark and DingTalk have four assets that their competitors do not possess:
- Organizational Relationships: DingTalk has 800 million users and a complete structure of 26 million companies (who reports to whom and what each department does); Lark has organizational data from large clients like Xiaomi and LiShi, which are unique corporate frameworks that new players cannot replicate.
- Permission Systems: The systems for determining who can view and modify what have been in place for years and are very mature.
- Historical Data: Messages, documents, and meeting minutes over the past decade represent a company’s “digital memory.”
- Trust and Compliance: Companies are willing to store their core data on these platforms, indicating trust in their security and compliance capabilities.
Are these assets a moat? Yes, because for AI to be useful, it needs to understand how companies operate, have the necessary permissions, and receive the data. However, they can also be a burden: these systems were designed for humans, and adapting them for AI requires significant rework. For example, the permission system, which was originally manual, must be automated for AI to function effectively.
3. ByteDance and Alibaba’s New Approaches: Consolidation vs. Diversification
Although their strategies differ, both aim to integrate AI capabilities with office scenarios:
- ByteDance: Consolidating into a “Triad”
Lark’s product team has been merged into DouBao (the AI team), and the GTM (marketing and sales) team has been merged into HuoShan Engine (cloud computing power). This combines the “office interface,” AI models, and computing resources to create a powerful system. The advantage is a closed loop where data, models, and computing power work together, allowing the AI to quickly learn from company needs; the downside is that if one component (e.g., the AI model) is weaker than competitors’ offerings, the entire system could be affected.
- Alibaba: Diversifying with a “Defense and Offense” Strategy
DingTalk continues to focus on its existing user base while QianWen Office, as an independent product, targets new users with a more lightweight AI approach. The advantage is that each component serves its purpose (DingTalk maintains users, and QianWen Office explores new directions); the downside is potential conflicts if they compete for resources and users.
4. The Key to Success: Practical Application and Commercialization
Regardless of the strategy, success depends on two factors:
- Whether the solutions make users dependent: For example, can AI automatically generate weekly reports, follow up with clients, or handle expense reports? The company that first creates such “killer features” will gain an advantage. Interviewees agree that integration with organizational structures is less important; what matters is whether the solutions truly solve problems.
- Finding new revenue models: In the past, Lark and DingTalk relied on per-user fees, but AI Agents aim to use fewer people to accomplish more tasks, which conflicts with this model. Alibaba’s QianWen Office still charges per user ($198/month), suggesting that the new model has not yet been fully tested. With low willingness to pay among domestic companies, whether a “pay-per-task” model will be accepted remains uncertain.
Conclusion: An Old War Ends, and a New One Begins
The decade-long competition between Lark and DingTalk over process standardization is over; the new battleground is personalized AI services. Both companies have their strengths and challenges. The winner will be the one that can quickly integrate AI capabilities into real business needs and find new ways to generate revenue. This new battle has just begun.