Chinese tech giants consolidate their AI agent portfolios, shifting toward the super workbench era.

Over the past month, Tencent, Alibaba, and ByteDance suddenly began doing the same thing.

They did not release new agents; instead, they began reducing the number of agents.

Tencent issued an internal notice: all business operations and certain teams from the Tencent QClaw Product Center have been integrated into WorkBuddy. WorkBuddy is touted as the third potential phenomenon-level product following QQ and WeChat—consider the weight of that assessment.

Alibaba has also moved, and with greater force. Caixin has obtained information that Alibaba is about to launch " Qwen "Office," integrating all three intelligent agent products—QoderWork, Wukong, and MuleRun—led by DingTalk’s new CEO, Chen Yusen.

In early July this year, these three products were just handed to him, and now they’ve become one. Qwen It is currently Alibaba’s most well-known AI brand, and integrating all office products into this system leaves no doubt about its intent.

ByteDance hasn’t been idle either. The AI programming product TRAE SOLO has quietly been renamed: TRAE Work. A single-word change has completely shifted its direction by 180 degrees.

Individually, each one is just a routine organizational adjustment.

Three major players almost simultaneously shifted their focus, funneling the multitude of agents they had aggressively promoted over the past six months into a single entry point—this is no coincidence.

This is the first true product consensus of the Agent era.

Recall what the scene was like at the very beginning of the Agent explosion?

Around the Spring Festival this year, nearly all companies felt that agents would be like WeChat official accounts back then.

One team does one, one department does one, one scenario does one. Everyone is betting heavily, but no one knows who will ultimately win.

In just a few months, thousands of agents with similar functionalities have emerged.

Technical barriers vanish instantly.

Tencent is a classic large-scale shrimp farmer. QClaw was developed by the Computer Manager team based on OpenClaw, WorkBuddy was developed by Tencent Cloud, and there are also QQ Shrimp and Browser Shrimp—different versions of Shrimp are scattered across various business groups, operating independently.

Alibaba is simultaneously running multiple Agent projects: the desktop AI agent tool QoderWork, which has remained in a semi-finished state and is continuously being refined; Wukong, which is also under ongoing development; and MuleRun, focused primarily on overseas markets. These products have similar positioning and heavily overlapping functionalities, causing computing resources and R&D efforts to be fragmented. As a result, users are confused and unable to form a unified understanding.

ByteDance hasn’t been left behind, successively launching several lobster agents—ArkClaw, ByteClaw, Feishu Aily, and Trae SOLO—deploying them like a strategic formation.

To be honest, this kind of concentrated betting is not unique to Chinese companies.

OpenAI, Anthropic, Google—all the same. Operators, Deep Research, Canvas, Projects, Codex, Claude Code, NotebookLM, Gemini… their product lines keep popping up like chimneys.

This is not a sudden phenomenon; when something new emerges during its exploratory phase, the best approach to management is to allow for repetition.

But experimentation requires spending money.

Over the past six months, major tech companies have heavily invested in agent development across multiple teams, consuming massive amounts of inference computing power. Multi-turn reasoning, repeated API calls, and redundant context retrieval have made the cost of a single agent execution far exceed that of standard conversational models. With costs remaining high and enterprise payment adoption lagging, fragmented product lines have significantly lowered the overall return on investment.

After open-source tools leveled the technological barriers, the core of competition changed. Computing budgets cannot sustain unlimited internal consumption; resources must be concentrated.

When the direction becomes clear, exploration ends.

This sounds contradictory.

Innovation is clearly a good thing—why cut it off yourself?

But if we look at the history of the internet over a longer period, this has happened over and over again. In fact, the entire internet has witnessed only three major attempts to establish a unified entry point.

In the PC era, browsers unified all web pages. In the mobile era, super apps unified all services. In the AI era, super workspaces are unifying all agents.

How many social products has Tencent developed?

WeChat, QQ, QQ Space, Renren, Tencent Weibo… in the end, WeChat consolidated everything and became the super gateway for mobile internet. Meituan started with group buying, movies, food delivery, travel and lodging, ride-hailing, and ultimately became a super app. Didi began with express rides, premium rides, carpooling, and chauffeur services, and eventually bundled them all into a single Didi app.

Each product consolidation is not due to failed innovation, but because the market gradually moves toward certainty. The true sign of a market's maturity has never been an increasing number of products, but a decreasing one.

In all platform wars, what ultimately matters is not the ability to create, but the ability to simplify.

Let’s take another look at how these three giants are clearing the field.

Tencent consolidated the exploration outcomes scattered across the Tencent PC Manager team into CSIG, which focuses on cloud services, integrated them into WorkBuddy, and transferred control to Tencent Cloud.

Alibaba has officially abandoned its decentralized approach of independently developing AI assistants across various business lines, and has now centralized control of office AI under DingTalk. Qwen Office and brand unified under one umbrella Qwen System.

ByteDove has shifted TRAE toward workflow collaboration, signaling the end of the era where agents operated as independent solo units—SOLO has become an invisible technical foundation within TRAE Work.

Throughout internet history, every product unification has signaled the beginning of a real war.

Beneath this consensus on consolidation lies a deeper shift.

Why must TRAE SOLO be renamed to TRAE Work? Why is Claude becoming less like a chatbot? What is OpenAI trying to achieve by aggressively integrating Chat, Code, Research, Projects, and Operator?

The answer is straightforward.

Last year, everyone thought the biggest market for AI was programmers.

Now, big companies are beginning to realize, no.

Programmers are just the first users of AI. The real market is everyone who works.

Undeniably, the first industry use case to be successfully implemented was coding. Tools like Cursor, Claude Code, and TRAE have all emerged. The reason is straightforward: code is highly standardized, operates within a digital feedback loop, and offers clear, precise error responses.

But writing code is just one part of the workflow. The scenarios that truly cover everyone involve checking emails, attending meetings, reviewing documents, processing data, approving requests, and following up on decisions.

The market for coding consists of tens of millions of developers. The market for general office productivity consists of billions of professionals. The scale and token consumption ceiling of the latter are orders of magnitude—hundreds to thousands of times—larger than that of the former.

Moreover, today's global giants are almost all converging in the same direction.

After encountering lukewarm reception for custom agents, OpenAI quickly shifted its focus to ChatGPT Projects and Operator, consolidating scattered features into a unified interface.

Microsoft has rearchitected Copilot, upgrading it from a standalone Office plugin to a central hub spanning the entire Microsoft 365 workflow.

Salesforce's Agentforce no longer emphasizes the flexibility of individual agents, but instead focuses on unified orchestration within enterprise CRM systems.

The domestic market is also accelerating. Tencent WorkBuddy, a strategic product with high expectations from Tencent, has rapidly increased its monthly active users (MAU) and daily active users in a short time, becoming the most active AI agent among domestic productivity applications.

Even traditional enterprise software and office giants can no longer sit idle; Kingsoft Office (WPS) recently launched similar products—the individual-focused Lingxi Professional Edition and the organization-focused WPS Comate—aiming to fully integrate AI agents into document and office workflows.

The gears of the business logic have thus completed their rotation.

A global consensus is emerging that as decentralized agent entry points gradually disappear, they are being replaced by super workspaces.

The three giants are clearly competing for dominance in the super workspace space, but don’t mistake this battle for a remake of traditional Office.

How to understand it?

Over the past two decades, enterprise software architectures operated in silos—ERP, CRM, OA, HR, finance, and project management systems each confined to their own domain, forcing employees to jump between them.

At this moment, it is people that connect everything. In the future, it will be agents that connect these systems.

This is also why Tencent, Alibaba, and ByteDance have independently entrusted this round of consolidation to their cloud and collaboration teams.

The execution efficiency of an agent does not depend on how intelligent the model itself is, but on how much enterprise-level data and API interfaces it can orchestrate.

Alibaba is about to launch Qwen DingTalk Office leverages DingTalk’s enterprise relationship chain, organizational structure, and approval workflows. Tencent’s WorkBuddy is backed by WeChat and Tencent Docs’ collaborative ecosystem. ByteDance’s TRAE Work is powered by Feishu’s knowledge base and workflow engine.

Big companies are consolidating agents because, at their core, they're fighting for absolute control over enterprise data and system APIs. Whoever becomes the first AI entry point employees open each day will control the central hub for all enterprise data and capabilities.

When users only need a single super workstation entry point, the remaining tens of thousands of software applications lose the reason to be opened directly.

Over the past year, a contentious view in the industry has been that AI is killing SaaS.

Many people believe that agents will come like a storm, completely overturning the ERP, CRM, HRM, and financial software that companies have used for over a decade.

In reality, no company dares to hand over its backend logic—built up over decades and tied to compliance and core assets—to be completely restructured by a large model.

In the past, the core premium of SaaS came from its user interface and workflow. Vendors meticulously designed every button, menu, and table, requiring employees to adapt to the software’s logic.

But when the Agent unified the workflow entry point, this entire set of rules was turned upside down. What truly changed everything was the widespread adoption of Skills (capabilities/interfaces).

The software no longer needs to present complex UIs to humans; it simply needs to connect its skills to the super workstation—whether it’s querying supply chain inventory in SAP, retrieving customer profiles in Salesforce, or generating financial reimbursement vouchers in Kingdee.

These capabilities, previously buried deep within dozens of submenus, have been packaged into standardized Skills.

This shift will directly restructure the industrial hierarchy of enterprise services: the front end will become a super workspace, the back end will consist of software, and the numerous vertical agents in between will have increasingly less space.

When employees stop accessing the SaaS interface, the per-user pricing premium will no longer work, and software vendors will have to shift from charging for UI access to charging based on the frequency of skill usage and delivery outcomes.

The greatest value of enterprise software over the past two decades has been its interface. The greatest value of agents is making the interface disappear.

And the bridge connecting them is Skills.

Looking back at the evolution of the Agent, the three stages are clearly evident.

In the first phase, the Agent is a product. Everyone is building it, with wildly varying forms and countless isolated systems.

In phase two, the agent becomes an entry point. Major companies are consolidating resources and fiercely competing to control the primary entry point for work and operating systems.

In phase three, an agent is a capability. After the entry point has been restructured, it is everywhere yet formless.

Today's consolidation efforts by Tencent, Alibaba, and ByteDance mark the industry's formal transition from the first phase to the second, rapidly moving toward the third.

Agents are experiencing the next wave of capability democratization in internet history, following browsers and super apps. They are transitioning from dazzling star products to essential, yet often unnoticed, foundational infrastructure—like electricity and the internet.

The agent won't become the new WeChat, but it's very likely to become the new Windows.

[Outside the layout]:

The development of technology always goes through an interesting process.

At first, it was something new. Then, it became a product. Later, it turned into a capability. Finally, it became nothing at all.

Electricity has no entry point, network protocols have no entry point, and databases have no entry point.

Their disappearance doesn't mean failure—it means they've become the foundation of the entire world.

Perhaps in a few years, we won’t discuss agents anymore—just as today, no one discusses HTTP.

Truly mature underlying technologies eventually lose their names.

This article is from the WeChat official account "Beyond the Layout," authored by HuaHua and BanJun.