Summary of Key Points
This in-depth industry analysis, spanning over ten thousand words, delves into the two most significant events in the recent domestic AI office sector: ByteDance's integration of the Lark and DouBao teams to launch the "DouBao Work" agent platform, and Tencent's release of the fully open WorkBuddy agent operating system. These two companies seem to have chosen completely opposite approaches—one focusing on "closed, deep integration" and the other on "building an open ecosystem"—but their ultimate goal is the same: to gain control over the future of corporate office operations. However, this control does not lie in the traditional notion of traffic through AI chat boxes. Instead, it revolves around the authority to entrust real work tasks and critical business data to AI, including the ability to execute tasks and verify results.
The article traces the evolution of agents from the first generation of robots 60 years ago to their current application in office settings, pointing out that the industry is still in its very early stages. Even a "unified browser" for agents has not yet emerged. The ultimate winner will not be the one with the most traffic or the highest model performance, but rather the one that can establish a corporate-level trust system that allows for error tracking, reversal of actions, and accountability.
---
Detailed and Accessible Explanation
1. Don't be misled by the "entry battle" mindset; what's at stake is trust, not just clicks
In the past, the competition focused on browsers and super apps on mobile internet, aiming to increase user clicks through pre-installations and subsidies to create impressive daily active user numbers. However, the logic of the AI office competition is entirely different. No company will entrust all their customer information, confidential contracts, or financial data to an AI just because its interface looks good or responds 2 seconds faster. Nor will they let it send emails, modify sales quotes, or handle payment approvals on their behalf. AI tools like Microsoft 365 Copilot were essentially assistants that followed human instructions; the final decisions and responsibilities always rested with humans. Now, the big players are competing for the "default authority to execute tasks." For example, you could simply say, "Generate a follow-up plan for customers who have been lost in East China for over three months," and the AI would handle the entire process—finding the relevant data, accessing the CRM system, writing the plan, and even initiating the approval process—all without your direct intervention. To gain this authority, traffic alone is insufficient; a comprehensive trust system is needed. This includes ensuring that AI understands what data it can access, that all actions are traceable, and that errors can be quickly corrected.
2. AI office agents have been in the works for 60 years, but they've always had limitations
Many think agents are a new concept brought about by large models. In fact, as early as 1966, Stanford developed the first robot, Shakey, capable of perceiving its environment, planning routes, and moving objects. However, for decades, agents were limited in their capabilities. The first limitation was their "action" aspect, with tools like macros and RPA that could automate repetitive tasks but could not understand natural language. The second limitation was their "comprehension" aspect; even with large models, they could only process text in chat boxes and could not directly interact with software or databases. It wasn't until 2022 that the combination of "reasoning" and "action" became feasible, enabling AI to actively use external tools and operate computers, thus truly becoming capable of working in office settings.
3. Tencent and ByteDance have chosen opposite approaches, but no one will win immediately
Some might think that the merger of Lark and DouBao by ByteDance, along with Tencent's open WorkBuddy platform, indicates a price war. However, the two companies are actually adopting different strategies. Tencent is building a "supermarket" approach, offering a platform where developers, industry experts, and hardware manufacturers can integrate their AI services into "Skills" (specific tools). For instance, a financial report analysis Skill could be used directly from the platform. The advantage of this approach is its broad range of capabilities, but the challenge is maintaining quality control, as any errors in these Skills could severely impact users. ByteDance, on the other hand, is creating a "boutique store" approach by integrating its existing enterprise systems (Lark’s document, calendar, meeting records, and customer data) with DouBao’s AI. This approach is more stable, as all data and permissions are managed internally, reducing the risk of data leaks. The challenge is integrating third-party services.
4. The winner won't be the one with the best model; workers need new skills
Currently, companies are competing based on model parameters and response speed, but these are not the most crucial factors in office scenarios. What matters is whether you can trust AI to handle your work. For example, you might not use an AI to generate financial reports if you fear it could alter the data. The ultimate winner will be the platform that can best manage errors, provide clear accountability, and ensure data security. For workers, skills like keyboard shortcuts and knowing how to use software hidden menus will become less valuable as AI becomes more advanced. Instead, the key skills will be clear communication of requirements and the ability to verify AI’s work. If you can't articulate your needs accurately or distinguish between good and bad AI outputs, you may soon be replaced by AI.
In summary, the competition in the AI office sector is about establishing trust and establishing reliable systems, not just about model performance. Workers need to develop new skills to work effectively with AI.