The Agent horse race has come to an end, and the super workstation has taken the leading position.

Over the past month, Tencent, Alibaba, and ByteDance have all suddenly embarked on the exact same initiative.

Rather than launching new Agents, they have started to scale back their existing Agent portfolios.

Tencent issued an internal notice, announcing that all business operations and partial teams from the QClaw Product Center will be merged into WorkBuddy. WorkBuddy is touted as a potential third phenomenal product following QQ and WeChat, a distinction whose significance you can fully appreciate.

Alibaba has also taken action, with even more aggressive moves. According to information obtained by *Caijing*, Alibaba is set to launch "Qianwen Office", integrating three existing agent products — QoderWork, Wukong, and MuleRun — into a single unified offering under the leadership of Chen Yusen, the newly appointed CEO of DingTalk.

Back in early July this year, these three products were transferred to his oversight, and in a very short time frame, they have been consolidated into one. As Alibaba's most well-known AI brand today, integrating all office products under the Qianwen system makes the company's intentions abundantly clear.

ByteDance has not stood idle either. Its AI programming product TRAE SOLO has quietly undergone a name change: TRAE Work. This single-word adjustment signals a complete 180-degree strategic shift.

Viewed in isolation, each of these developments appears to be a routine organizational adjustment.

However, the fact that three major tech giants are acting almost simultaneously to converge the large number of Agents they aggressively promoted over the past six months into a single unified entry point cannot possibly be a coincidence.

This represents the first genuine product consensus in the Agent era.

I. We No Longer Need So Many Agents

Let's recall the unprecedented boom that marked the initial explosive phase of Agent development.

Around this year's Spring Festival, nearly every company believed that Agents would replicate the success trajectory of WeChat Official Accounts in years past.

Every team would build one, every department would build one, and every use case would have its own dedicated Agent. Companies were placing bets left and right, with no one certain which product would ultimately emerge victorious.

In just a few short months, thousands upon thousands of agents with nearly identical functionalities flooded the market.

Technological barriers were instantly reduced to zero.

Tencent was a typical example of this widespread "mass rearing" approach. QClaw was developed by the PC Manager team based on OpenClaw, while WorkBuddy was built by Tencent Cloud. In addition, there were QQ Lobster, Browser Lobster, and various other "Lobster" variants scattered across different business groups, all operating independently with no unified coordination.

Alibaba simultaneously ran multiple Agent projects internally: QoderWork, a desktop-level AI agent tool that remained perpetually in a semi-finished state; Wukong, which was constantly being refined; and MuleRun, which was focused on overseas markets. These products had overlapping positioning and largely redundant functionalities, spreading out computing power and R&D resources, confusing users, and preventing the establishment of a unified public perception.

ByteDance was no exception, launching several Lobster Agents in succession — ArkClaw, ByteClaw, Feishu aily, and Trae SOLO — deploying them across the market in a broad, sweeping formation.

Frankly speaking, this kind of intensive, scattered betting is not a unique characteristic of Chinese companies.

OpenAI, Anthropic, and Google have all followed the exact same pattern. Operator, Deep Research, Canvas, Projects, Codex, Claude Code, NotebookLM, Gemini... their product lines have sprung up like isolated chimneys, one after another.

This is not a sudden, unprecedented phenomenon. When a new technology first emerges during its exploratory phase, the best management approach is to allow for a certain degree of redundant experimentation.

But trial and error comes at a significant financial cost.

Over the past six months, the major tech giants running multiple parallel Agent lines have consumed massive amounts of inference computing power. Multi-step reasoning, repeated API calls, and redundant context retrieval have pushed the operating cost of a single Agent run far higher than that of a standard conversational model. While costs have remained persistently high, enterprise-side adoption and paid services have not kept pace, and the fragmented product lines have dragged down the overall return on investment to very low levels.

After open-source tools eliminated technical barriers, the core of competition fundamentally shifted. Computing power budgets can no longer support unlimited internal waste, and resources must be consolidated.

Once the path forward becomes clear, the exploratory phase naturally comes to an end.

II. Major Tech Giants Are Proactively Phasing Out Their Own Innovations

This statement may sound contradictory at first glance.

If innovation is inherently a positive force, why would companies deliberately choose to shut down their own innovative projects?

But if we look back across the broader history of the internet, this pattern has played out over and over again. In fact, the entire internet industry has only witnessed three major eras of unified entry point consolidation.

In the PC era, browsers unified access to all web pages. In the mobile era, super apps unified access to all services. Now, in the AI era, super workstations are in the process of unifying all Agents.

How many social media products has Tencent launched over the years?

WeChat, QQ, Qzone, Pengyou, Tencent Weibo... and in the end, a single WeChat brought everything under one umbrella, becoming the super entry point for the entire mobile internet. Meituan expanded into group buying, movie ticketing, food delivery, travel and hospitality, and ride-hailing, eventually consolidating all these services into a single super app. Didi launched express rides, premium rides, carpooling, designated driving services, and ultimately integrated all of them into one unified Didi platform.

Every instance of product consolidation does not happen because innovation has failed. It occurs as the market gradually moves toward greater certainty. The true marker of a mature market is never an ever-increasing number of products, but rather a steadily decreasing number of them.

In all platform wars, the ultimate competition is not about the ability to create new things, but about the ability to make deliberate, strategic reductions.

Let's take a closer look at how these three tech giants are clearing up the market landscape.

Tencent has brought the exploratory achievements scattered across the PC Manager team back under CSIG, the business group focused on cloud services, integrating them into WorkBuddy and placing full operational control under Tencent Cloud.

Alibaba has officially abandoned its previous decentralized approach of allowing individual business lines to freely incubate their own AI assistants. Full authority over office AI development has been transferred to the DingTalk-led Qianwen Office project, with all related branding fully integrated into the Qianwen ecosystem.

ByteDance has reoriented TRAE toward workflow collaboration, effectively bringing an end to the era of Agents operating as standalone, independent tools. The SOLO functionality has been transformed into an invisible, underlying technical foundation embedded within TRAE Work.

Throughout internet history, every large-scale product unification has signaled the official start of the real, high-stakes competitive battle.

III. Programmers Are No Longer the Largest Target Market

Beneath this industry-wide consensus on consolidation lies an even more profound strategic shift.

Why was it so necessary for TRAE SOLO to be renamed TRAE Work? Why is Claude increasingly moving away from being a simple chatbot? Why is OpenAI integrating Chat, Code, Research, Projects, and Operator with such relentless urgency? What exactly are they aiming to achieve?

The answer is very straightforward.

Over the past year, everyone assumed that programmers represented the largest market for AI.

Now, the major tech giants are beginning to realize that this assumption was incorrect.

Programmers are merely the first wave of AI users. The truly massive market consists of every single person who goes to work.

It is undeniable that the coding use case was the first to achieve widespread commercial viability in the industry. Tools like Cursor, Claude Code, and TRAE all gained significant traction for a simple reason: code is highly standardized, operates within a fully digital closed loop, and offers clear, unambiguous feedback for error correction.

But writing code is just one single step within a much larger end-to-end workflow. The use case that truly covers every professional worker includes checking emails, attending meetings, reviewing documents, processing data, approving requests, and tracking decision-making progress.

The coding market is measured in tens of millions of developers. The general office productivity market encompasses billions of working professionals. The latter's scale and maximum potential for token consumption are hundreds, even thousands of times larger than the former.

Furthermore, nearly all the global tech giants are now converging on this exact same strategic direction.

After its custom agent initiative underperformed expectations, OpenAI rapidly shifted its focus to ChatGPT Projects and Operator, consolidating its previously scattered functionalities back into a single unified interface.

Microsoft has restructured its entire Copilot architecture, upgrading it from a standalone Office plugin to a super entry point that runs through the entire Microsoft 365 workflow.

Salesforce's Agentforce no longer emphasizes the flexibility of individual agents; instead, it highlights the unified orchestration capabilities built directly into the enterprise CRM system.

The domestic Chinese market is also accelerating this transition. Tencent WorkBuddy, a strategically prioritized product with high expectations from the company, has seen its monthly active users (MAU) and daily active users surge rapidly in a short period of time, making it the most active efficiency-focused AI agent application in China.

Even traditional enterprise service and office software giants can no longer afford to stay on the sidelines. Kingsoft Office (WPS) recently launched products with a similar positioning: Lingxi Professional Edition for individual users, and WPS Comate for organizational clients, aiming to fully integrate AI agents into document and office scenarios.

The gears of commercial logic have thus completed their full rotation.

A global consensus is taking shape: as the independent entry points for scattered, standalone Agents gradually disappear, they will be replaced by unified super workstations.

IV. What Is Being Fundamentally Rewritten Is the Very Nature of Work

The three major giants are clearly competing to dominate the super workstation market, but this battle must not be misunderstood as a mere updated version of traditional Office software.

How can we properly understand this?

Over the past two decades, enterprise software architectures have operated in silos: ERP, CRM, OA, HR, finance, project management — each system guarded its own narrow scope of functions, forcing employees to constantly jump back and forth between different platforms.

During that era, human users were the ones connecting all these disparate systems. In the future, Agents will be the ones that tie all these systems together.

This is precisely why Tencent, Alibaba, and ByteDance have all, almost simultaneously, placed this round of Agent consolidation under the purview of their cloud and collaborative office teams.

The execution efficiency of an Agent does not depend on how "intelligent" the underlying model is; it depends on how much enterprise-level underlying data and API interfaces it is able to orchestrate and leverage.

Alibaba's upcoming Qianwen Office is backed by DingTalk's massive enterprise relationship network, organizational structure data, and pre-built approval workflows. Tencent's WorkBuddy draws on the synergistic ecosystem of WeChat and Tencent Docs. ByteDance's TRAE Work is supported by Feishu's extensive knowledge base and mature workflow engine.

By consolidating their Agent portfolios, these tech giants are essentially fighting for absolute control over the orchestration of enterprise data and system APIs. The company that becomes the first AI entry point employees open every single day will control the central orchestration hub for all enterprise data and capabilities.

Once users only need a single super workstation entry point, thousands of other software applications will lose their reason to be opened directly.

V. Software Is Beginning to Move Into the Background

Over the past year, a highly debated viewpoint has circulated throughout the industry: AI is killing SaaS.

Many people believe that Agents will act like a sudden storm, completely overturning and rebuilding the ERP, CRM, HRM, and financial software systems that enterprises have relied on for over a decade.

But the reality is that no enterprise would ever dare to hand over the entire backend logic it has built up over more than a decade — logic that is tightly tied to compliance requirements and core business assets — to a large language model for a full rewrite.

Traditionally, the core premium value of SaaS came from its interactive interface and user-defined workflows. Vendors meticulously designed every button, menu, and form, forcing employees to adapt their own work patterns to the logic of the software.

But once Agents unify the primary work entry point, this entire set of long-established rules is completely upended. What is truly driving this transformation is the widespread adoption of Skills (functional capability interfaces).

Software no longer needs to display complex user interfaces to human users. All it needs to do is connect its Skills to the super workstation, whether that means enabling supply chain inventory queries in SAP, customer profile retrieval in Salesforce, or financial expense report generation in Kingdee.

These capabilities that were once buried deep within dozens of nested sub-menus are now packaged into standardized, easily accessible Skills.

This transformation will directly restructure the industrial hierarchy of the entire enterprise services sector. The frontend layer will belong to super workstations, the backend layer will belong to specialized software providers, and the market space for a large number of vertical, niche Agents will continue to shrink.

Once employees no longer open traditional SaaS interfaces, the per-user licensing premium tied to the UI will lose its validity, forcing software vendors to shift their revenue model away from selling access to a graphical interface and toward charging based on Skill invocation frequency and the actual value of delivered outcomes.

Over the past two decades, the most valuable component of enterprise software has been its user interface. The greatest value of Agents is to make that interface disappear entirely.

And the bridge that connects these two worlds is none other than Skills.

VI. The Best Agent Is the One You Never Notice

If we look back at the evolutionary trajectory of Agents, it clearly unfolds across three distinct phases.

In the first phase, the Agent itself is a standalone product. Everyone is building them, they take wildly varied forms, and isolated, siloed products spring up everywhere.

In the second phase, the Agent becomes a unified entry point. Major tech giants begin to consolidate resources, competing fiercely to seize the primary entry points for work and operating systems.

In the third phase, the Agent evolves into a ubiquitous capability. Once the entry point reconstruction is complete, it is everywhere, yet completely invisible and intangible.

The ongoing consolidation efforts by Tencent, Alibaba, and ByteDance mark the official transition of the industry from the first phase into the second, as it rapidly accelerates toward the third phase.

Agents are now undergoing the same kind of capability democratization that the browser and super app brought to the internet in previous eras. They are evolving from a highly celebrated, star product into a foundational infrastructure as indispensable and unnoticeable as electricity or internet connectivity.

Agents will never become the next WeChat, but they have every chance of evolving into the next Windows.

A note beyond the main text:

The development of technology always follows a fascinating, predictable trajectory.

In the very beginning, it is a novel curiosity. Then, it matures into a distinct product. Later, it becomes a widely adopted capability. And in the end, it simply ceases to be seen as something separate.

Electricity does not have its own dedicated entry point. Internet protocols do not have their own dedicated entry point. Databases do not have their own dedicated entry point.

Their "disappearance" does not represent failure. It means they have become the fundamental underlying infrastructure that the entire world runs on.

Perhaps in a few years' time, we will no longer be having conversations about Agents, much like no one today holds discussions about the HTTP protocol.

Every truly mature underlying technology eventually loses its distinct brand identity and simply becomes part of the fabric of everyday life.

This article originates from the WeChat Official Account "Beyond the Layout", authored by Huahua and Banjun, and is republished by 36Kr with official authorization.