Over the past two weeks, I have felt several times that it is not necessary to continue researching AI-powered office work at present.
It's not because it's unimportant, but rather because it's too important and too lively. Tencent, Alibaba, ByteDance, Baidu, and Kingsoft have almost all made their moves, with new products and organizational adjustments being announced one after another, along with rounds of advertising investments. The media and analysts are also keeping pace.
As of today, there are at least five things that we think have been clearly analyzed:
Tech giants have fully entered the AI-powered office era. This is not just a product trial by a single company, but a collective reallocation of resources.
Agents are shifting from "answering questions" to "completing tasks". Writing documents and making PPTs is just the surface, what's more important is breaking down tasks, invoking tools, and executing across systems.
Beyond models, data, context, permissions, and workflows are becoming new competitive variables.
Feishu, DingTalk, WPS, and Enterprise WeChat, these traditional office assets, may regain value due to AI.
The basic unit of office software is also changing. In the past, sales were based on files, functions, and seats, but now more and more people are discussing tasks, outcomes, and even digital employees.
For those looking to understand how far this war has reached, there are already several worthwhile articles to read.
For example:
Coverage has tracked the trend to two paths: standalone agents and legacy office workflows, as well as the migration of office products from software features to task delivery.
Economic Observer: With internal competition suspended, Tencent, Alibaba, and ByteDance's AI office products are joining forces to form an alliance.
Industry Insider｜"Dou Bao Work" In-Depth Analysis of Feishu: AI Office Officially Enters the "Context" Battlefield - Bringing Enterprise Context, Permissions, and Organizational Data to the Forefront of Competition
The Interface/JMedia: "The Great Shift in AI Offices: Who Will Deliver Real Results First" - the question has been further pushed from "who has more functions" to who can truly embed themselves in workflows and deliver results.
These outstanding achievements have laid the foundation for this research.
If simply re-proving these conclusions, then launching another round of research would be meaningless. However, upon further examination, there are still some unclear points that gradually coalesce into three additional questions:
Why does China's internet industry engage in an all-out war every few years, a battle that no one can afford to sit out?
2. As intelligence becomes a public supply, where is the new scarcity?
3. When all capabilities are on an equal footing, what can possibly become the differentiating Power?
Why does China's internet industry always seem to be on the verge of another major battle?
If we stretch our memory a bit, the scene we are seeing today is not unfamiliar.
The hundred-team battle, the ride-hailing war, the food delivery war, shared bicycles, community group buying, and now the ongoing competition for C-end AI assistants - almost every few years, China's internet sees a new battlefield emerge, with capital, giants, and startups all rushing in at the same time.
The six wars are certainly not the same. Especially for C-end AI assistants, the outcome is still far from determined. Putting them together is not to summarize a "law of internet wars", but to establish a set of historical contrasts: when a new market has just been proven to be viable, and a large amount of capital enters the market at the same time, which actions will repeatedly appear? What will happen to the things that are initially the most scarce and valuable?
What is truly worth taking a closer look at are three of them.
In the food delivery wars, capital was indeed burned on massive subsidies, but what remained was not just user numbers. The riders, merchants, order density, delivery scheduling, and city fulfillment gradually formed a real-world production system. The money ultimately turned into a capability that is difficult to replicate in just a few months.
Shared bicycles are a case in point. Capital quickly turned "whether there are bikes on the street" from a scarcity to a surplus, but the bikes themselves did not automatically become a long-term competitive advantage. The more bikes there are, the more important maintenance, scheduling, wear and tear, and capital occupation become.
Retaining assets does not necessarily mean retaining long-term competitive advantages.
Community buying has provided another reminder that having capital, traffic, and experience in waging large-scale battles, as Didi does, is not enough to automatically translate into procurement, warehousing, and supply chain capabilities in the fresh retail sector.
When the battlefield changes, what was once the most valuable may no longer be worth as much.
These wars, when considered together, make me more concerned about the recurring movement behind them.
At the outset of many wars, capital chases the scarcest resource: if there aren't enough users, buy users; if there aren't enough drivers, subsidize drivers; if there aren't enough vehicles, flood the city with cars.
But capital is not just a bystander that only discovers scarcity.
Money, once it flows in on a large scale, creates its own supply.
As what was once scarce becomes increasingly common, the value of mere ownership begins to erode, and competition shifts to the next bottleneck that is harder to replicate.
What is happening here is actually a kind of scarcity migration: as capital and technology continue to expand the originally scarce supply, old advantages begin to depreciate, and competition shifts to the next more difficult-to-replicate bottleneck.
This reminds me of an issue we've been studying recently - the AI capital cycle.
The AI capital cycle is a capital-centric perspective on our long-term research into the AI industry. It examines how capital chases scarcity, expands supply, drives bottleneck migration, and ultimately alters costs, competition, and value distribution. It not only tracks where money flows into models, computing power, applications, and infrastructure, but also investigates how these investments are converted into revenue, cash flow, and long-term control, and who ultimately bears the risk.
There's an observation that I increasingly feel is suitable for explaining today's office wars:
Capital always chases scarcity, but it itself constantly expands supply, pushing scarcity to the next link.
This process can be roughly divided into five steps:
The past few internet wars have only given us a set of benchmark rates, rather than answers.
What's really making today's AI office warfare interesting is that the same process may be happening to something that was rarely expanded like this in the past:
Intelligence itself.
Models were once extremely scarce. Training a sufficiently good large model was itself the ticket to entry.
However, over the past few years, capital, computing power, talent, and technological advancements have all converged on this field. Models have become increasingly numerous, powerful, and affordable.
If the scarcity migration continues, the issue will no longer be just "who currently has better AI capabilities."
The more pertinent question is:
As intelligence itself is increasingly supplied, where will the next scarcity in AI competition migrate to?
And from here on, this war took a truly divergent path from the internet wars we are familiar with.
2. As intelligence starts to become a public supply, where is the new scarcity?
The real divergence occurred in the concept of "intelligence" itself.
In past internet wars, capital could certainly be used to subsidize supply, but many core capabilities still had to be built brick by brick by the companies themselves.
You can subsidize riders, but you can't buy a nationwide delivery network that's already running smoothly in a single quarter. You can pour money into ads to attract users, but you can't simply acquire another company's supply chain, merchant relationships, and city-level operational experience along the way.
Large models have changed this premise to some extent.
Intelligence is becoming a producible resource that can be purchased
An AI application company does not need to have the world's best foundational model to be qualified to develop very good products.
The model can be invoked through API calls, switched between different tasks, and even combined with multiple models in the same product. Today, there are already several AI office products that allow users or systems to select different models based on tasks.
This is not just a change in technical architecture, it has also changed a very fundamental competitive condition:
Intelligence is increasingly becoming a kind of productive resource that can be externally procured and called upon as needed, rather than just a capability exclusively owned by a particular company.
In the past, a food delivery platform without a delivery network was nothing.
Today, an AI application without the strongest model can still directly purchase intelligent capabilities that are fairly close.
This is also why the migration of scarcity in the AI era may be happening faster than in the past.
In the past internet wars, capital could make traffic a standard configuration.
In the Agent era, capital and technological advancements may even make intelligence a standard feature together.
Of course, this does not mean that models are unimportant, nor does it mean that all models will become the same.
Frontier models still determine the upper limit of capabilities and may also determine the product experience for a period of time. However, for application companies, the question has changed: how long can model leadership be monopolized?
If competitors don't have to retrain your model from scratch and can simply purchase an API or switch to a better foundation, the competitive advantage established solely through model capabilities will have a much shorter time window than in the past.
What is truly changing is the "scarce resource" that capital is competing for and constantly eliminating.
02｜For the first time, application capabilities will be proactively "depreciated" by progress at the underlying level
AI-powered offices are particularly sensitive to this change.
Compiling dozens of documents, generating a presentation deck, analyzing spreadsheets, operating a browser, and completing tasks across multiple software programs—these are, of course, genuine competitive skills today.
But how much of that will still constitute a difference after the next round of model upgrades?
This is a very special risk of AI application.
A delivery network that an internet company painstakingly built over the years won't suddenly depreciate just because the operating system releases a new version next month.
AI products can.
Functions that required specialized teams and complex engineering to achieve yesterday may become public configurations directly generated by the underlying model tomorrow, thanks to upgrades in reasoning capabilities, multimodal support, tool invocation, or long-context capabilities.
The company's own products have not deteriorated.
The entire industry's floor has suddenly been raised.
This is a kind of capability depreciation that is easily overlooked: not server depreciation, nor amortization on financial statements, but rather the original scarce application capabilities being continuously commoditized by advances in underlying technologies.
So I am increasingly willing to use a simpler sentence to judge AI applications:
Model upgrades continuously raise the capability floor, and applications must build Power above the floor.
What this statement truly requires application companies to answer is not "whether models are still important," but rather:
As more and more basic capabilities become publicly available, what is left for you to own?
A new competitive landscape is also beginning to emerge here.
An English term is increasingly being used in the industry: harness (Agent runtime support layer). It can be roughly understood as the system outside of the model that enables the Agent to work stably, including tools, context, task status, permissions, and verification, among other things.
The more a model resembles a public good, the easier it is for competition to migrate to these areas.
Harness is also not a new universal moat, if the capabilities within it can also be purchased, accessed, or migrated away, scarcity will continue to decline.
03｜Even existing office assets are transitioning from "exclusive use" to "on-demand"
This brings the issue to one of the most compelling perspectives in today's AI office competition:
Everyone can have models, but Feishu has documents and organizational relationships, DingTalk has enterprise systems, WPS has office documents, and Enterprise WeChat has enterprise collaboration and the WeChat ecosystem.
These legacy assets certainly matter.
If AI is to eventually enter real-world work, it cannot stay confined to an isolated chat box forever.
But another thing that has happened over the past two years is that having old office assets no longer naturally means winning.
Agent is working to connect tools and data that were originally closed off within different software.
MCP - Model Context Protocol, a model context protocol - is one of the most typical examples. It attempts to provide a standard way for Agents to call external data and tools after obtaining authorization.
MCP itself is not the main character in this war.
What truly matters is the direction it represents:
Agents do not necessarily need to own all the factors of production themselves, but can also use them through interfaces.
Such attempts can already be seen in China. For example, Qianwen Office can connect to external services like DingTalk, Feishu, and Microsoft 365 through Connector. What's most noteworthy here is not the specific types of files it can search or the number of functions it can call, but rather that it proves one thing:
A new AI office product does not necessarily have to replicate an entire set of traditional office software in order to be qualified to enter that work environment.
This is a new issue for all the old platforms.
Much of the value of many software assets in the past came from a very simple phrase:
Others can't reach it.
And now, it may be gradually becoming:
Others can use it, but they have to go through my interface, my identity, and my rules.
If you want to see what this looks like taken a step further, Microsoft is a good reference point.
What Microsoft 365 has accumulated over decades goes far beyond Word, Excel, and email—it also encompasses emails, meetings, Teams chats, files, people relationships, and records of organizational collaboration.
Microsoft refers to the ability to understand these work information as Work IQ.
If following traditional platform logic, this should be locked within its own products as much as possible. However, Microsoft has chosen to further open up Work IQ as an API, allowing third-party agents to call work context within Microsoft 365 within existing identity and permission boundaries.
Microsoft has not abandoned its old assets, but is redefining how to derive value from them.
The ultimate agent that acts on behalf of users may not necessarily come from Microsoft, but it needs to understand emails, meetings, files, and organizational relationships, and may still go through Microsoft's systems.
This is slightly different from the past platform logic of "the entrance must be completely mine".
Traditional office platforms may not only compete for the final chat window or agent, but also retreat to a more fundamental level, becoming the foundation for work context, system records, identity permissions, and even governance infrastructure.
But this cannot be taken to the other extreme either.
Being interconnected does not mean being interchangeable.
An external agent being able to read a Feishu document does not mean it immediately inherits the organizational relationships, collaboration history, and permission structure within Feishu.
Being able to access Microsoft 365's data does not mean that third-party products have taken Microsoft's position in the enterprise IT system.
The interface solves the issue of "whether it can be used".
It has yet to answer who is more knowledgeable, who is more trustworthy, who can preserve work state over the long term, who controls authorization, and where responsibility falls when something goes wrong.
This is also where this round of chip scarcity migration truly starts to get interesting.
On the one hand, the supply of models is increasing, and many application capabilities are being rapidly commoditized.
On the other hand, the data, tools, and some context that were originally confined to office software have also begun to become producible factors that can be authorized to be called.
The model may belong to one company, the agent to another, the documents and emails may be stored in a third company's system, and the identity and access permissions may be controlled by the enterprise's own IT system.
Things that were previously tied to a single app are being dismantled layer by layer.
So at this point, simply comparing models, functions, and user bases, or even who has the most complete office ecosystem, is no longer enough.
What truly needs to be answered next is:
When models can be invoked, tools can be connected, and even part of the data and work context can be authorized for use, where has the new scarcity shifted to?
When all capabilities are being caught up with, what can possibly become the differentiator?
It is here that Power needs to be formally brought out.
The Power referred to in this article is not just a general term for "advantage" or being slightly ahead today. It is closer to a structural control power, meaning that even if others can do something similar to you, it is still difficult to bypass you, and you can even influence the rules by which others choose and act.
What it cares about is whether newly emerging scarcity can crystallize into long-term value and control.
How to determine if an advantage has reached this point?
A very practical stress test is:
If a competitor replicates 90% of your overt capabilities, why can't users easily leave you?
This is the Power Test used in the article. Instead of creating another comparison table of functions for Tencent, Alibaba, ByteDance, Baidu, and Kingsoft, it's better to put the assets that are often considered as "moats" today into the same set of questions.
First, let's downplay a few advantages that seem significant at first glance
This Power Test only asks five questions:
Is this thing difficult to rapidly increase supply for?
Can competitors obtain it through procurement, API, or Connector?
How high are the migration costs when users want to move away?
Will it continue to compound as it is used?
In this table, model capability and feature leadership should be the first to be downgraded.
The model initially fails to pass this round of testing: it still determines the qualification to go public, but as an independent scarcity that is being continuously supplied, it is difficult to explain why its value ultimately remains.
Functionality is also similar, with today's stunning PPT, Deep Research, or browser operations possibly becoming the industry's minimum configuration soon.
Traffic patterns are more complex. Tencent has WeChat and QQ, ByteDance has Douyin and Doupak, Alibaba has consumer and enterprise ecosystems, and Baidu still holds search, browsers, and traditional information portals. These distribution capabilities can determine who sees a new product first.
However, the cost of switching between AI products is not high. A user can simultaneously use Doupai, DeepSeek, Yuanbao, and Kimi, with switching costs far lower than relocating an entire social network or a company's business system. Traffic remains important, but it's more like controlling the initial entry point. After the first task is completed, the real challenge begins: why do users come back for the second and third time?
02｜Old Office Assets Are Starting to Become Another Problem
Looking further down, old assets like Feishu, DingTalk, Enterprise WeChat, and WPS are obviously much more robust in terms of product functionality.
They have documentation, meetings, organizational relationships, enterprise customers, a permission system, and workflows.
But the previous question has already shifted the issue to another location:
The data can still reside in your system, but the tasks may be completed by someone else's AI work system.
So the most intuitive judgment of the past — "I have data you don't, therefore I have a moat" — now demands one more question:
Can others use my data?
If the answer shifts from "completely inaccessible" to "can be used, but must be authorized," scarcity hasn't disappeared, it has just been pushed back.
Microsoft provides a comparable example overseas: it allows external agents to call work contexts while keeping identity, permissions, compliance, and governance at a lower control level. The opening up of old assets does not necessarily weaken their value, but rather may upgrade them from being exclusive to the front-end to becoming a system record layer or control layer.
Therefore, having data does not equate to having the power to control how it is used.
Feishu is no different. Just because an external AI can read a document doesn't mean it suddenly has access to the entire organization; the extent to which an agent can act on a user's behalf remains constrained by identity and pre-existing authorization boundaries.
This has pushed the issue further back again:
Having privileges is not equivalent to controlling who can obtain those privileges.
Agents can obtain significant execution permissions, but the true decision-makers on the origin of these permissions, when to revoke them, whether they can be used across systems, and who is responsible in case of issues, often remain the enterprise's own identity system, administrators, and governance rules.
So, "old office assets will be revalued" is not a universal truth.
Some assets may genuinely appreciate in value because they have secured a new control point.
Some assets may simply retreat from their former position as the sole front-facing interface, becoming raw materials that any Agent can invoke with authorization.
The key is no longer how much you have accumulated in the past, but where these things are positioned in the new technological structure.
03｜An old asset may lose its exclusivity while gaining new value at the same time
The same asset may experience two opposing events simultaneously during the Agent era.
On the one hand, open interfaces make it increasingly easy for others to call upon it, thereby decreasing the exclusive value of "only I can use it".
On the other hand, if more and more AI systems have to go through it, it may instead become a more underlying and stable infrastructure.
Microsoft's Work IQ is one possibility.
The enterprise identity and permission system is also a possibility.
Even office systems like Feishu and DingTalk may not necessarily need to rely on "ultimately owning that AI chat window" to prove their value in the future.
The real question is:
Will legacy assets in the new technological architecture devolve into raw materials that anyone can summon at will, or will they be upgraded into a control layer that more and more agents cannot bypass?
This question is obviously more difficult than "who originally had an office ecosystem".
It also explains why it's impossible to simply stand on the balance sheet of any existing giant to predict the outcome.
Having office software doesn't guarantee victory.
Not having a full suite of legacy office software doesn't mean you're out of the game.
Because the Agent is rearranging the positions of these assets relative to each other.
Models, AI office portals, documents, and business data, as well as identity and access permissions, may all belong to different systems. What users ultimately see is the result of a task, but behind the scenes, it is no longer a traditional product chain completely owned by a single company.
Power Test has gotten to this point, having at least ruled out several overly early answers.
The strongest model alone isn't enough.
The largest traffic volume is still not enough.
Having the most functions is not enough.
I already have a full set of office software, but it's still not enough.
These are either rapidly scaling up supply, becoming connectable, or controlling only one layer of the entire task chain.
If they really want to continue searching for even stricter controls, they will have to look even further down.
Pursue those things that are difficult to completely obtain through a single API call, difficult to buy out with one round of financing, and difficult to move together after a user clicks "export" once.
In addition to context, there is also the state of work
When discussing AI-powered offices today, almost everyone will mention the importance of context. This is undoubtedly true, as an AI work system that is unaware of who you are, the project background, what meetings have been held in the past, and where files are stored will find it difficult to truly work on your behalf.
However, the context is still too broad.
Knowing what happened in the past and knowing the current progress of a task are two different things.
The former is closer to Context—the contextual meaning.
The latter is closer to State.
For example, I ask an AI office system to help me research a company's competitive landscape. It has read 20 documents from yesterday and knows my research topic—that's the context.
But if I come back the next day, it still knows which materials have been excluded, which judgment I rejected, which numbers are still pending verification, why the draft stopped here, and what should be supplemented next under normal circumstances, only then can it begin to approach a working state.
The difference between the two is somewhat like that between a person who has just joined a project and one who has been working on it for three months. The former can review all the meeting minutes again, while the latter understands why things have developed to this point.
Microsoft's Work IQ has begun to address this issue by building it into the infrastructure, providing Agents with real-time work context and Workspaces where long-running Agents can store data, files, memories, progress, and intermediate results, eliminating the need to start from scratch every time.
What is surfacing here may be another kind of scarcity:
Work continuity.
In the past, office software emphasized "all data is here".
Agentic's work system may be more worthy of attention:
Work has been happening here.
Data can be exported, APIs can be opened, and third-party calls can be authorized.
But one thing is that why it has been done up to this point, which options have been ruled out, what corrections have users made, and who the project is waiting for - these states continue to accumulate, and it's not necessarily possible to completely move them with just one connection.
It is only at this level that the concept of "using it for a longer time makes it more valuable" truly begins to emerge.
A genuine AI office product is not equal to a single agent, as a product can simultaneously encompass a main entrance, workspace, orchestration layer, multiple agents, skills, tools, and different models, with the agent that executes specific tasks and the underlying model being interchangeable.
So the real question worth asking is not:
Are users always assigning tasks to the same Agent?
Instead,
Users tend to default to assigning a certain type of task to a particular system over the long term.
Can that system preserve state, carry over feedback, coordinate multiple agents, and pick up the same task when the next one arrives?
"Continuous Authorization" is a strong candidate, but not yet the answer
If this relationship holds over the long term, what the AI work system accumulates will no longer be just a single visit.
As tasks such as meeting preparation, project follow-up, industry monitoring, and report updates continue to occur within the same system, it may become increasingly familiar with the user's work status, error correction records, historical choices, tool combinations, and authorization boundaries.
This has a very intuitive difference from the entrance competition in the mobile internet era.
The dispute used to be over:
I'll start with WeChat.
In the Agentic work environment, competition may gradually turn into:
When something comes up, which system do I default to handling it first?
"Opening up" is mainly a matter of attention.
It then begins to approach a commission relationship.
It takes time to form and may accumulate with use, and can further affect the routing of tasks: the ultimate decision of which model, which tool, and which SaaS a task calls upon is not necessarily made by the user every time.
At this point, continuous commissioning is a strong Power candidate, but existing evidence is still insufficient to directly define it as the final Power of the Agent era.
The reason lies in the fact that another force is also strengthening at the same time.
MCP is reducing the cost of having Agents individually adapted to different tools and data sources, while A2A is attempting to enable Agents from different manufacturers and frameworks to communicate and collaborate with each other. The two sets of open protocols solve different problems, but both aim to improve the composability of the Agent system.
Meanwhile, enterprises are also tightening control over identity and permissions.
Microsoft can now establish an independent identity for the Agent and manage its authentication, authorization, scope of permissions, lifecycle, and auditing with Entra. What the Agent can ultimately access depends not only on its model capabilities, but also on the identity and authorization granted to it by the enterprise.
These forces combined will dismantle many traditional forms of lock-in.
Companies can switch out front-end AI products while still retaining their original Microsoft 365 data, Feishu work assets, and IAM systems; a work system can also break down tasks to be completed by multiple different agents.
If long-term states themselves gradually become more transferable in the future, the lock-in strength of "default delegation" could continue to weaken.
So power may not be entirely concentrated in a single new super app.
It is more likely to remain dispersed across several layers over the long term: models supply the intelligence, the agent and orchestration layer understands and coordinates tasks, work systems maintain state and records, the identity and permissions layer determines who can do what, and enterprises retain ultimate governance for themselves.
These layers will compete with each other and also be open to one another.
The AI office software battle appears to resemble the next super entrance war, but it may ultimately not produce a new super entrance.
It's not that the competition is not fierce enough, but rather that the intelligence, tools, data, status, identity, and execution rights that were previously tied to a single app are now being unbundled into different layers.
From the perspective of the AI capital cycle, this path becomes clearer: capital chases scarcity and continually expands existing supply, driving scarcity to migrate. Model capabilities are being rapidly supplied, and old office assets are becoming increasingly callable. Next, what's truly worth pursuing is what will continue to accumulate, become increasingly difficult to migrate, and ultimately gain the ability to control the next step.
September 2026, where has this war gone?
By September 2026, China's AI office landscape is no longer about six companies competing on the same track by comparing functions, but rather they are exploring various paths to form different kinds of Power.
Tencent has reached a very interesting position.
Starting in late August, WorkBuddy began noticeably strengthening its "memory and evolution" capabilities, long-conversation continuation, and session-state preservation. By September 2, Tencent had pushed this approach a step further, officially launching the WorkBuddy open platform to make underlying Agent capabilities available to smart-hardware makers, industry applications, and developers, with more than 100 ecosystem partners onboarded in the first batch.
This means Tencent is currently trying two things simultaneously: on the one hand, it is further integrating its vast connections and distribution network into WorkBuddy, allowing it to remember "where a task was left off last time"; on the other hand, it is attempting to turn its Agent runtime platform into a foundational infrastructure that other applications can also use. If these two paths converge, Tencent will be competing for not just an AI-powered office entry point, but also for ongoing commissioned relationships and the Agent runtime layer itself.
2. Ali is a different story.
DingTalk has long accumulated organizational structures, approvals, workflows, and extensive enterprise software relationships. Qianwen Office now integrates Skills, MCP, connectors, local files, browsers, and scheduled tasks into the mix. Notably, scheduled tasks can operate independently of a user's immediate query, continuously invoking files, skills, and connectors to complete work. What Alibaba is attempting is to transform existing organizational processes from "data an Agent can read" into "work methods an Agent can continuously execute." The next question is whether these processes can further accumulate state and grant Qianwen genuine task routing authority.
Thirdly, what's most noteworthy about ByteDance is that it made a choice in terms of its organization.
In late August, the TRAE and Docket-related teams were further integrated into the Doubao system, and the productivity routes of Feishu and Doubao were also converging, after which "Doubao Work" was officially launched.
The personal AI portal and enterprise collaboration assets have been clearly integrated into the same main pipeline for the first time. However, just because organizations are merged does not mean that users' work relationships have naturally merged. What ByteDance is really betting on now is: can the personal AI relationships established by Douyin cross over into the organizational work relationships carried by Feishu.
Kingsoft is directly turning one of the aforementioned hypotheses into a product.
WPS began enhancing its project space, document space, and long-term tasks in August, allowing data, conversations, and cloud documents to be continuously retained around the same project. In the past, WPS's greatest asset was "files are always here", but now it is trying to take a step forward - "projects are always here and continue". The truly valuable metric going forward is not how many project spaces are built, but whether this status and modification history will ultimately form migration costs.
5. Baidu, on the other hand, is the most suitable for conducting a reverse test.
On August 27, Baidu's DuerOS launched 15 industry suites and 96 professional skills, clearly focusing on professional fields such as finance, business, law, product development, operations, and human resources. Unlike Feishu, DingTalk, and WPS, which have a thick traditional office foundation, DuerOS bypassed the direct competition of "recreating Office" and instead focused on professional work methods and task delivery quality. If this approach succeeds, it will prove that traditional office assets are not the only ticket to entering the next-generation work system.
6. If this route is further pushed towards the AI-native end, Kimi is a more typical example.
Kimi Work is not inheriting a traditional Office empire, but is instead growing its own project folders, conversation branches, Agent browser, runtime permissions, skills, and plugin systems. It is attempting to build a work environment from the inside out, starting with task execution. The most noteworthy aspect is whether these native Agent capabilities can ultimately settle into a stable state and become default delegations, rather than just a set of useful but easily replicable features that can be overtaken by the next generation of models.
So as of September 2026, at least three distinct paths to power in China's AI office space are being tested simultaneously: some are turning legacy office assets into agents, others are pushing super-entrances deeper into the workflow, and still others are trying to build new working relationships from the ground up by executing AI-native tasks. Players like ByteDance are even straddling two paths at once.
No single approach has yet proven itself the certain winner.
As they move forward, they are increasingly colliding with the same few scarce resources: working states, continuous commissions, identity permissions, and routing rights that determine where models, tools, and services are directed.
This sector has already entered a scale of hundreds of millions of users. According to QuestMobile statistics, in July 2026, the overall MAU of the AI efficiency office sector was approximately 102 million, with the number of uses growing 112.4% year-over-year. This indicates that it is no longer just a niche for tech enthusiasts. However, scale can only tell us who occupies the entrance faster; an increase in usage frequency alone is still not enough to prove who has formed a dominant position.
The real turning point may only emerge when these users start leaving their jobs behind.
From now on, we will focus on five signals:
Will a long-term working state create real migration costs. When switching to a new AI work system, how much will be lost in terms of project status, historical choices, error correction records, and work habits.
To what extent will third-party AI access legacy office assets, merely reading them or going further to frequently write, execute transactions, or even bypass the original office product interfaces?
Who ultimately gets the real routing power. After users hand over tasks, do models, tools, and SaaS still need to choose for themselves every time?
Where does control over identity, permissions, and governance ultimately reside? AI work systems, office platforms, enterprise IAM, or the enterprise itself?
Will the value-capture model keep migrating? Will enterprise payments move beyond seats and software subscriptions toward tasks, outcomes, agent calls, work status, or governance capabilities?
Any significant change in any of these factors may force us to modify our judgment today.
So this is still just a phased study at this point in September 2026.
If you're working on AI-powered offices, corporate agents, or enterprise IT, I'm particularly interested in knowing: what things that seem "rentable" or "migratable" today are actually impossible to move in real-world enterprises; and what things that are currently considered moats are being circumvented.
We especially welcome materials that prove this article's judgment was wrong.
(End)