This week, another debate broke out in our editorial department. The focus of the argument was: in the battle for AI office applications, what exactly is the decisive factor?
I believe that the foundation model is the most important.
The reason is that in industry groups and exchanges with practitioners, people generally feel that the interfaces and features of AI office products are largely similar. What truly determines whether to pay is whether it can get the job done well. This depends on the model's capabilities, as well as its integration with other office software and workflows.
For example, many AI office software applications are willing to integrate Kimi because of its stronger context capabilities; the combination of Doubao Work and Feishu also has the magical effect of "truly understanding office workers."
However, another colleague countered me, arguing that channels and marketing are the trump card for AI office applications.
A friend working in an enterprise told him that many people in the company are using WorkBuddy, and they don't even know that Doubao and Qwen also have AI office products. This can only be explained by the one-click connection via WeChat, which provides a convenient entry point, coupled with recent aggressive promotional discounts, generous free quotas, and low usage costs.
To put it bluntly, most office tasks in ordinary enterprises are not that complex or technical. Who reaches users first through marketing campaigns is the decisive factor.
This time, neither side convinced the other. When we zoom out, we find that everyone is not even paying attention to the same game.
The foundation model serves as the defensive line for AI office applications, underpinning product capabilities; battling for users' attention through marketing and channels is the most intense midfield scramble for the ball. Meanwhile, the forward line in industry scenarios such as finance, banking, and government affairs requires a decisive strike deeply integrated into specific business operations.
A broader arena means that a single point of advantage cannot directly win the game. The competition in AI office applications is far more exciting and much longer than any of us imagined.
The Midfield Battle for AI Office: Fierce Marketing Combat
In this debate, our conclusions come from different research subjects. One conclusion comes from observing professional AI users and media practitioners, while the other judgment comes from communicating with ordinary enterprise users.
Although everyone has different understandings of AI and different identities, they share a commonality: they all feel that AI office products are homogenized, but they also think that product similarity is not a big deal.
What professional users care about is the delivery quality of AI office products and whether the job is done well. Therefore, the product form is not important, and capabilities are largely similar; integration with the office ecosystem is more important.
For example, Qwen integrates with DingTalk, Doubao Work relies on Feishu, and WorkBuddy combines with WeChat. Wherever your previous work results and communication documents are deposited, you will find the AI office product integrated with it more user-friendly and more understanding of you.
From the perspective of ordinary enterprise users, as long as a product can already meet their basic needs, there is little motivation to compare it with other similar products.
Even if there are model differences and functional distinctions between different products, it is hard for them to feel obvious improvements in experience and benefits. Even if there are, they are not necessarily willing to switch products for this reason.
As AI office products become increasingly similar and fail to form clear differentiation, the natural competition among vendors turns into vying for more exposure and brand awareness.
Consequently, channels and marketing have come to dominate the midfield battle for AI office applications. Since 2026, all players have been actively burning money to buy traffic, and the marketing war has become fiercely intense.
According to media estimates, the overall offline advertising cost for WorkBuddy in just the four cities of Beijing, Shanghai, Guangzhou, and Shenzhen is close to 100 million CNY. After the launch of Doubao Professional Edition in June, the advertisement "New work habits, let Doubao do it first" was massively deployed in major airports, striving to plant a mental anchor for business travelers. Qwen Office, together with 7 partners, has also covered over 60,000 high-end office buildings nationwide.
Doubao, WorkBuddy, and Qwen have therefore been pushed to the center of the table, becoming the current "Big Three" in AI office applications.
But here comes another question: if marketing can really settle the matter once and for all, why are products like DuMate, Kimi, and OfficeAce still entering the market intensively? Blanket burning of money to grab entry points can temporarily seize a midfield advantage, but it cannot win the game in one go.
To understand the core of competition in AI office applications, we still need to shift our focus to the forward line and the defensive line.
The Defensive Line: The Model War
Among deep AI users and industry media groups, everyone is highly familiar with large models and AI applications at home and abroad, and switching between various products is also very frequent. What they care more about is whether the underlying models can provide sufficiently advanced intelligence for business operations.
The foundation model, a corner that the general public cannot see and does not care about, is precisely the defensive line and the rear guard for AI office applications.
After the release of Kimi K3 in July, Tencent announced that the built-in models of the entire WorkBuddy product line support Kimi K3. Prior to this, WorkBuddy had already built in a batch of mainstream models such as Hunyuan, DeepSeek, GLM, and MiniMax. Overseas, Cursor also opened the model option for Kimi K3 at the end of July.
Kimi's ability to seize a place in the AI office market relies precisely on the technical barriers brought by its foundation model.
To complete a complex task, office agents often require the collaborative execution of multiple agents, continuously passing task progress across multiple stages, such as in-depth research on dozens of documents, writing long reports of hundreds of pages, or analyzing complex projects and comparing product plans of different versions. Whether AI can deliver completely depends on its long-context capabilities.
The context window of Kimi K3 supports up to 1 million tokens. The larger the context window, the more documents the model can process in a single task and the longer project background it can retain, which is particularly attractive for these complex office tasks.
This is exactly where the value of the foundation model lies. The competitiveness of AI office applications does not necessarily require burning massive amounts of money to monopolize entry points; it can also serve as a capability base that other products are willing to call upon.
For products like WorkBuddy and Cursor, if they want to deliver a great office experience, training a flagship foundation model from scratch is not the only way. Introducing advantageous external supplies like Kimi at the model layer can enhance product competitiveness and hold the capability defensive line for AI office applications.
Kimi is not an isolated case.
The desktop rankings of the AI Product Chart (AICPB) for July 2026 show that the total monthly active users for domestic AI office desktop applications are about 30 million. WorkBuddy ranks first with 11.1523 million, and DuMate ranks second with 6.7430 million. Among them, DuMate's month-on-month growth reached 1063.79%, which is even higher than WorkBuddy's 304.40%.
If we only look at the midfield marketing war, DuMate is obviously not the player with the loudest voice. Its growth obviously cannot be simply attributed to marketing.
On the one hand, Baidu's previous accumulation of AI products among government and enterprise customers may have provided DuMate with a starting point to enter office agents, without having to fight for users from scratch. On the other hand, Baidu's accumulation in foundation models and technology also supports DuMate's capabilities, enabling it to continuously execute complex tasks across applications and files.
The cases of Kimi and DuMate illustrate that the competition in AI office applications does not have just one battlefield.
Relying on mass traffic and marketing to break through is one approach; in the defensive line, relying on the foundation model to continuously raise the capability ceiling of office agents, thereby seizing the growth opportunities of office intelligence, can also hold a territory in AI office applications.
The Forward Line: The Decisive Strike Deep into Industries
Of course, even if everyone can call upon powerful foundation models, it does not mean that each can get the job done well.
Because foundation models provide general capabilities, while business rules and workflows vary greatly across different industries and positions. This requires AI office applications to delve into specific businesses and transform general capabilities into functional details adapted to industry needs.
Just as the ordinary enterprise user mentioned at the beginning said, he knows nothing about various models and has no interest in understanding all AI office products on the market one by one. He only cares about whether the application he happens to use is user-friendly.
Industry scenarios are the inevitable path for AI office products to establish differentiation, and also the forward line of the game.
Take an example, mainstream AI office products all emphasize their long-horizon task capabilities with multi-step and multi-agent collaboration. However, when it comes to specific positions, such as a researcher and a finance professional, although both need this capability, the requirements for product details are vastly different.
Writing research reports requires reading materials in various formats such as web pages, PDFs, and PPTs. This requires office agents to be able to recognize heterogeneous data in various formats and cross-validate them.
In financial work, financial Excel files for multiple months are often placed in the workspace. Although the data structures are similar, the templates from different departments may vary, and the names for the same expense may differ. This requires office agents to follow strict business rules and standard accounting subject definitions for understanding and alignment.
The error tolerance rates for these two types of tasks also vary. Whether to introduce manual review in the workflow and at what stage to introduce it is highly specific, which cannot be handled by simple professional templates alone. The forward line battle, with a deep understanding of industry scenarios, provides opportunities for more AI office products to enter the field.
For example, the recently upgraded Huawei Cloud OfficeAce emphasizes professionalism in serious scenarios and comes with over 20 preset expert agents. This upgrade further expands relevant skills in the finance field, providing capabilities such as compliance verification and auditing.
The productization and specialization of Huawei Cloud OfficeAce also demonstrate that organizing general model capabilities into industry workflows and delivering results according to real business standards is precisely the value of the forward line.
Facing the ultimate industry users, whether they find AI office products user-friendly could also be the decisive strike that determines the winner.
Before the Final Whistle
Looking at the whole game, the AI office competition is far more exciting than imagined.
The midfield marketing war, burning money to grab users, is the most lively with fierce competition. But outside the spotlight, foundation model vendors in the defensive line are continuously expanding the boundaries of intelligence, refreshing our imagination of AI capabilities time and again. In the forward line, products that deeply understand industries are also striving to make AI understand us better, enabling it not only to do the work but to do it well.
The midfield battle is increasingly difficult to constitute a barrier on its own, and perhaps it will become less and less important in this AI office war.
There is still too much room for exploration in the forward line and the defensive line. Therefore, the story of AI office applications is far from the final whistle.