Summary of Key Points
Office Agents (AI-powered office intelligent systems) have become a new battleground for giants such as Alibaba, Tencent, and ByteDance. However, smaller companies like Zhipu, Yuezhi Dianmian, and MiniMax have not joined in. The reason is not a lack of desire, but rather the inherent differences in their capabilities, scale, and the complexities of providing enterprise-level services. The advantage of the giants does not lie in their stronger models (in fact, some smaller companies' models have performed better in technical tests); it lies in their control over the organizational context (permissions, approval processes, and process data) and the Runtime (the system that enables the AI to interact with the enterprise environment). DeepSeek, on the other hand, has chosen an open-source approach, entering the market by leveraging a public platform. With the market growing rapidly, the giants' strategic moves are both about capturing new opportunities and defending their existing positions in the B2B sector (with platforms like DingTalk and Lark). The smaller companies are either developing personal desktop tools or betting on open-source standards; they have made their choices but have not directly entered the giants' competitive arena.
Why Can Giants Dominate Office Agents While Smaller Companies Lag Behind? The Difference Lies in “Organization” and “Infrastructure”
Office Agents generate revenue from enterprise purchases, with high and stable average transaction values, which is certainly attractive to smaller companies. However, there are two main reasons why they cannot keep up:
- Lack of Organizational Context: Enterprise office applications are not just about individual software use; they need to be integrated into organizational structures. Who has access to financial data? How do reimbursement processes work? This type of information is stored on the servers of platforms like DingTalk, Lark, and WeCom, and smaller companies cannot obtain it to train their models. The giants' agents are “enterprise employees” that can operate within the existing permission systems, while smaller companies' tools (such as Kimi Work) are merely “personal assistants” that cannot manage organizational-level tasks.
- Inability to Handle Complex Operations: Enterprise procurement processes are lengthy, ranging from business requirement formulation, IT selection, and POC (Proof of Concept) validation (2 to 1 month), to bidding, legal contract review, and information security testing, all of which may require the CEO's signature, potentially taking up to half a year. After a purchase, a dedicated team (2-5 people) is needed to maintain the system and adapt it to changes in the organizational structure and custom requirements. Smaller companies, with more unstable cash flows, cannot afford such time-consuming processes.
Office Agents Are Not Just Chatbots; They Are “AI Employees with a Physical Presence”
Many people think of office agents as chatbots, but in reality, they are combinations of “brain” and “body”:
- The Model is the Brain: It understands instructions (e.g., “organize last week’s meeting minutes and send them to the boss”), but without the “body” (the runtime system), it is useless. The model can only think; it cannot access files or send emails.
- The Runtime is the Body: It provides the necessary functionality to turn ideas into actions. For example, when the model says “read this file,” the runtime system opens the file and extracts the content; when it says “send a weekly report,” it connects to the email system to send it. This cycle of “thinking → acting → receiving feedback → thinking again” is the core of the runtime system.
The giants' runtime systems are extensions of the enterprise organization (e.g., Qianwen Office uses DingTalk’s enterprise capabilities, while DouBao Work uses cloud computing services), serving the entire company. Smaller companies' runtime systems, on the other hand, are extensions of the individual user (e.g., Kimi Work uses a local execution kernel), serving the individual user.
Is the Market About to Explode? Giants Are Spending Money to Secure Market Entry Points to Defend Their Positions
The growth of the office agent market is astonishing:
- In June, the number of visits to 17 domestic desktop AI office agents exceeded 60 million (compared to 20 million in March), with Tencent WorkBuddy topping the list with 20.97 million visits.
- The enterprise-level AI agent market is expected to reach 44.9 billion by 2026 (from 21.2 billion in 2025) and could exceed 332 billion by 2029.
- Gartner predicts that 40% of enterprise applications will incorporate AI agents by 2026.
Why are the giants so eager? Because the way users interact with these agents is changing. If users get used to speaking directly to the agents without using platforms like DingTalk and Lark, the organizational assets that the giants have built over the years could depreciate. Therefore, their efforts are both offensive (to capture new markets) and defensive (to protect their existing customer bases). For example, DouBao Work offers a 30-day subscription free of charge when users sign up, and Qianwen Office offers a public beta trial. WorkBuddy provides tokens as a incentive.
DeepSeek’s Unique Approach: Focusing on Open-Source Standards Instead of Developing Complete Products
DeepSeek stands out among the smaller companies. It does not develop a complete office agent but instead opens-source the tool used to create agents (DeepSeek Harness):
- This tool allows users to add custom functions (such as image recognition plugins). The GitHub page received over 20,000 star ratings within an hour of being released and over 180,000 in 9 days, making it the fastest-growing open-source project in history.
- Its goal is not to make money from enterprises but to become a common foundation for office agents, similar to how the Android system is a standard for mobile devices. However, this approach also carries risks: the MIT license allows anyone to modify the code, and users could switch to other models like GLM or GPT, potentially leaving DeepSeek dependent on default API calls for revenue generation, which could put it at a disadvantage.
Smaller Companies Are Not Giving Up: They Are Choosing Different Paths
Smaller companies are not giving up on office agents; they are choosing approaches that suit them better:
- Developing Personal Desktop Tools: Companies like Kimi Work and Zhipu AutoClaw focus on personal office tasks (organizing files, writing reports) and avoid the complexity of enterprise-level processes, allowing for quicker revenue generation.
- Betting on the Open-Source Ecosystem: DeepSeek uses its tool to attract developers. While it may not generate immediate profits, it could become a standard in the long term, allowing for revenue through API fees or ecosystem sharing.
For smaller companies, there is no need to defend existing customer bases or spend heavily on market entry points. They are exploring their own paths in a market where office agents are still in their infancy, and the future may not be limited to the role of “enterprise employees” alone.
Conclusion
The battle for office agents is just beginning. Giants are leveraging their organizational strengths and runtime systems to capture the enterprise market, while smaller companies are seeking differentiation through technology or open-source strategies. In the future offices, agents will become a new entry point, but who will win depends on who can truly address users' core needs—whether it’s improving organizational efficiency or enhancing personal productivity.