Lobster Carcasses, SaaS Pipe Dreams, and Big Tech Anxiety: What Exactly Is AI Office Work?

In 2026, AI office software has become a business that every major tech company must pursue.

Three months after its release, Tencent's WorkBuddy reached 20.97 million monthly PC visits; Alibaba integrated its three Agent products—QoderWork, Wukong, and MuleRun—into "Qwen Office"; ByteDance's "Doubao Work" attracted applications from over 100,000 enterprises for its beta within a week of launch.

For a time, AI office ads were everywhere—in airports, subways, and elevators. Open Xiaohongshu or Bilibili, and you'd find AI office tutorials and recommendations overflowing the screen. If you don't normally follow AI, the flood of advertising might leave you completely baffled about what's happening. But at the very least, everyone would form the same first impression: AI office is definitely a good business—otherwise the major tech companies wouldn't be entering the fray so densely, wouldn't be investing so many resources.

But is that really the case? I want to pour some cold water on this: tracing the origins of the AI office boom reveals a business lacking commercial prospects. The major companies' collective entry is less a proactive choice than being forced onto the table. Driven by both external trends and internal pressure, AI office is likely to become a desperate last stand with a despairing outcome.

Everything starts with "picking up the corpse of the lobster."

Lobster Carcass

At the beginning of this year, Lobster set off a wave of intelligent agents worldwide. This cross-application, execution-heavy application model logically fits the narrative standards of AI applications perfectly, and in practice it genuinely delivers considerable office enhancements—for a time, it inspired boundless imagination.

On the other hand, Lobster's security is questionable, and its installation and usage experience is overly complex, dooming this open-source application to be unable to occupy the commercial mainstream long-term. When Lobster itself is kicked out of the game by various forces, who will fill the market space it has already opened up?

What lies on this old path? There's real Token consumption—in March of this year, Lobster's Token call volume once accounted for 18% globally, genuinely proving that customers are willing to burn money on intelligent agents. There's also real application growth—Lobster doesn't rely on models but conquered users with the agent application model. This showed the major companies, exhausted from battling over foundation models, a brand-new direction for AI development. At the same time, Lobster naturally possesses a certain ecosystem affinity—during the phase when everyone was raising lobsters, users went to great lengths to make Lobster compatible with WeChat, Feishu, and DingTalk. So if these vendors launched their own Lobster-like AI office applications, wouldn't they naturally plug into their own moats?

Only one problem remains: Lobster is an open-source project, and it hasn't proven any commercial potential or built any payment habit scheme.

In other words, Lobster makes absolutely no money. And the major companies, who must make money, have no model they can directly copy.

And so, the major companies picked up Lobster's carcass. More precisely, they picked up the shell Lobster shed. This shell looks glossy and splendid, with scales and claws flying—but the problem is, the shrimp meat called money doesn't grow on the shell.

So where does it grow? Nobody knows the answer.

SaaS Residual Dream

A well-known data institution, in its enterprise software report, characterized China's enterprise services market as a "SaaS desert."

Cruelly, after more than two decades of development, China has essentially proven that SaaS is an utter failure. Enterprises would rather pay high costs for computing power and privatized solutions than pay even a tiny subscription fee for software. Every round of enterprise software upgrades tells SaaS a new story. Subscription models, seat-based pricing, traffic-based billing, mutual conversion between free and paid versions, privatized delivery for large clients. Each time, the story sounds exciting, but when it comes to charging, there's nothing to be done.

Unfortunately, the cutting-edge AI office and Token billing seem essentially to be just such a new SaaS story. In this story, software seems able to transcend applications, do work on its own, perform natural language understanding—but it still lacks the answer to that question: why should users keep paying?

The twenty-year predicament of Chinese SaaS can be condensed into one sentence: the value of software is hard to perceive, and the habit of paying has never been cultivated. The essence lies in the fact that China's internet never lacks free things; the small and medium enterprises that form the backbone of the economy don't care about efficiency, only about revenue; large enterprises trust localization more and are unwilling to be bound by paid software. These fundamentals remain fully intact in the AI era. AI office software can help employees work faster—so what? Will investors pay for this and let everyone leave work early?

So the change Lobster brought occurred on the product and technology side. This product model, validated as having a market when free and open-source. But on the commercial side, it may be a completely different situation. All the residual dreams of SaaS are still reflected in AI office. There's no new driver for enterprise payment, no strong factor to change payment habits. The moment you mention payment, there's nothing.

1. There must be a policy of exchanging free access for scale, exchanging subsidies for users, and the competition is fierce. Then tiered charging becomes the standardized path, and it's also discovered that funnel-type payment doesn't make much commercial splash.

English text is being prepared.

3. High-frequency homogeneous competition will rapidly squeeze profit margins. Major companies still have to return to the old path of relying on ecosystem positioning, while smaller players will quickly be eliminated. The big fish will eat the small fish, but the meal will be very painful.

High penetration, low payment, low gross margin, free subsidies in exchange for scale, yet scale cannot be converted into money. These SaaS lessons are very likely to repeat themselves in AI office.

But its means and methods are currently unknown to us all.

Anxiety of major companies

Whether it is policy, public opinion, or the capital market, all are pushing major companies to do AI. Major companies have indeed invested sky-high AI costs, including computing power procurement. But after the investment comes the need for output, and as a result the major companies have found that AI, a money-devouring beast, is very hard to find customers to take it on.

The situation in the C-end market is many users but no money. AI users' monthly active users have already exceeded half of China's internet users, rapidly completing scale popularization. But the free model makes it very hard for AI to monetize directly. The vast majority of users simply do not want to pay for AI, and vendors even have to continuously pay the inference costs of models. This loss-making business is unprecedented in the internet era, and on top of that, model differentiation is narrowing, making competition among vendors even fiercer. And the few high-net-worth, high-tech-demand users are flocking to overseas products such as GPT and Claude Code.

In this situation of "the house is finally renovated, only to find it leaks rain from all sides," the patience of the capital market has begun to narrow, and major companies must actively seek cash flow that can be injected. As a result, the B-end market has become almost the only certain way out. After all, B-end cash flow is predictable, measurable, and renewable, and enterprise migration costs are extremely high, with user stickiness far higher than that of individuals.

Old wine in a new bottle

The reason is that what AI office brings is mostly office efficiency improvement, and employees no longer need to do a lot of complex and redundant work. For example, the same document written with AI, the same spreadsheet pulled faster by AI. In theory this direction has extremely high industrial value, but in practice the commercial ceiling is very low. Because the payer for efficiency improvement is very vague: the direct beneficiaries are employees, but the one who pays is the boss. And the boss finds it very hard to pay for employees working faster, especially on the basis that he is not used to paying for software and cannot find more work for employees to do.

Individual efficiency improvement does not equal organizational efficiency improvement, and enterprises will not be willing to pay quantitatively and continuously for individual efficiency. And making individual users pay for AI office must again face problems such as the limited willingness and ability of individuals to pay. To be honest, many of the AI office users I have come into contact with have a certain willingness to pay. But when paying, they are more inclined toward overseas products with better results, and more people are willing to pay to try something new rather than include it in continuous expenditure.

If AI office wants to gain sustained vitality, the most ideal way out is to create new value for enterprises rather than simply "improving office efficiency." Being able to bring output that did not originally exist, helping enterprises earn more money, deliver more orders, and save entire job lines—these are the anchors for sustainable charging. But the lobster has not brought these values, so the new wine is still waiting to be brewed.

AI office is probably just like this—the shell shed by the lobster has constructed its product form. This shell is very beautiful, but inside it is empty, with no business model left. The basic model of SaaS forms its commercial core, which is a residual dream of twenty years in which 99% of enterprises lost money. But even if the first two both hold, major companies still must bet on AI office. Because anxiety makes it impossible for major companies to let go of any opportunity to recover funds; even if there is no cash in this opportunity, they still have to look for whether there is a ticket to the next opportunity.

If within the next year AI office cannot achieve a sufficient payment rate, a relatively high proportion of outcome-based billing revenue, and a scaled continuously paying user base, then that vicious cycle of being praised but not selling well will play out again. And the real watershed is probably still whether AI office moves toward "saving people trouble" or "producing output for enterprises."

The shell shed by the lobster has already been put on by major companies, but flesh will not grow inside this shell by itself.

What major companies really need to go toward are the places where flesh and blood can grow.

This article comes from the WeChat public account "Brain Polar Body" (ID: unity007), author: Feng Ciyuan, published by 36Kr with authorization.