Exiting after just 188 days, big-company AI products are accelerating their reshuffle

A "lobster" is preparing to exit, and the next round of Personal Agent competition has already begun.

On September 24, Tencent QClaw announced that due to business development adjustments, it will cease operations on December 24, 2026, and will provide tools to help users migrate their data to WorkBuddy, another Tencent product. From its public beta on March 20 to the announcement of its service shutdown arrangements, only 188 days passed for QClaw.

On the other side, major tech companies are busy laying out the new track. On October 13, Alibaba Qwen announced the upcoming release of the personal assistant product "Qwen Go!", attempting to help users handle matters such as flight delay compensation claims, shopping price comparisons, and phone appointments. Previously, ByteDance was also reported to be advancing a personal assistant product, with the related project once using the codename "Spell".

Such scenes have recurred repeatedly over the past four years. We reviewed the product moves of Tencent, Alibaba, Baidu, ByteDance, as well as Moonshot AI, Zhipu, MiniMax, and DeepSeek—eight companies—across ten major AI tracks.

The data presents two core characteristics: first, competition cycles are shortening extremely fast, from text-to-image and ChatBot to video generation and AI Agent, with new trends following one after another; second, product lifecycles are getting shorter, with renaming, integration, and shutdowns becoming the norm, and standalone products continuously converging toward general assistants and underlying platforms.

Four years, ten tracks: more and more trends, and major companies following up faster and faster

In 2022, text-to-image tools such as Midjourney and Stable Diffusion attracted attention, and "input a sentence, generate an image" became the starting point for many people to understand generative AI. Domestic companies quickly followed up, with products such as Baidu's Wenxin Yige and Alibaba's Tongyi Wanxiang launched one after another, and AI formally entered scenarios such as painting, design, and content creation.

Subsequently, large language model ChatGPT pushed this wave to a broader population. In 2023, products such as Baidu's Wenxin Yiyan, Alibaba's Tongyi Qianwen, ByteDance's Doubao, and Moonshot AI's Kimi appeared intensively. The chat box gradually became the basic form of AI applications, and different needs such as search, office work, and companionship also unfolded around it.

By 2024, the industry was exploring in two directions simultaneously: on one side, multimodal video generation, with products such as ByteDance's Jimeng, Kuaishou's Kling, Zhipu's Qingying, and MiniMax's Hailuo Video appearing one after another; on the other side, AI hardware, with companies such as Alibaba, Baidu, and ByteDance attempting to put intelligent assistants into devices such as earphones and AI glasses, striving for user entry points beyond mobile apps.

Taking the "lobster" as an example, from January to March 2026, various "lobsters" arrived one after another. In March alone, five companies—ByteDance, Tencent, Zhipu, Baidu, and Alibaba—concentratedly disclosed their own "Claw-type" products. Compared with the long months experienced in the early period from ChatGPT's release to domestic large models following up, there is now very little observation time left for companies with new hot spots.

These waves are not simply "replacing one another in sequence". While new directions keep emerging, investment in old directions continues. What companies face is no longer just finding the next trend, but deciding amid ever-stacking layouts: which products are worth continued investment, which capabilities can be merged, and which experiments should end.

Major companies want "all of it", startups "bet on priorities"

Among the ten tracks counted this time, Alibaba and ByteDance cover all ten, while Tencent and Baidu each cover nine. In contrast, among the four emerging AI companies, Moonshot AI covers six, Zhipu five, MiniMax four, and DeepSeek only two.

Behind this difference are differences in companies' resource endowments and business priorities. For internet majors, AI is not only about opening new businesses, but also a tool for transforming their original ecosystems. Whether search, content, office work, e-commerce, or hardware, their huge existing users and business scenarios provide them with ample testing grounds and also give them more markets they need to defend.

After the tracks are laid out, which products remain on the field?

Entering a track does not mean being able to continuously maintain a standalone product. As can be seen from the statistical chart, after the tracks were laid out, each company has been continuously adjusting product forms and whether products stay or go.

Basic entry points are retained by all, but other tracks are frequently reshuffled.

All eight companies have laid out chat assistants, which are the basic entry point for search, creation, and task processing. But in other directions, product exits and shutdowns occur from time to time.

Taking AI companionship as an example, over the past four years, MiniMax's Glow was taken down, Moonshot AI's Ohai stopped updates, Tencent Music's Weiban and Baidu's Yuexia ceased operations. But within the same track, products such as Talkie, Xingye, and Maoxiang still operate normally.

After casting a wide net, major companies generally move toward internal integration and consolidation.

ByteDance is characterized by a broad layout while continuously adjusting the division of labor among products. Doubao, Jimeng, and Maoxiang respectively take on different needs, the Xinghui team was merged into Doubao, the MarsCode programming assistant was changed to a TRAE plugin, and the TRAE team was later also merged into the Doubao system...

Some of Baidu's capabilities are consolidated into existing systems. Wenxin Yige was moved into Wenxin Yiyan; Wenxin Kuaima was merged into Baidu Dazi; the chat assistant underwent renaming and entry migration from "Wenxin Yiyan—Wenxiaoyan—Wenxin". While continuing to participate in multiple directions, it is reducing some standalone entry points.

Alibaba and Tencent also have integration phenomena. Alibaba's "Wukong" was integrated into Qianwen Office in July 2026; Tencent's QClaw has announced that it will cease service in December 2026 and guide users to migrate to WorkBuddy.

The paths of startups have diverged.

DeepSeek's consumer product layout is relatively concentrated. Compared with other companies continuously launching new names and new entry points, it mainly adds capabilities around intelligent assistants. Zhipu's text-to-video product "Qingying" was merged into "Zhipu Qingyan"; MiniMax's general assistant "Hailuo AI" was later renamed MiniMax, and the original brand shifted to the video business.

Product shutdowns or consolidations involve not only changes in user demand and intensifying market competition, but also stem from internal resource trade-offs within companies. The layout at launch is only a starting point; only after sustained investment can a company decide which products deserve to exist independently.

The trajectories of six typical products: renaming, business divestiture, and consolidation

Observing the iteration trajectories of six typical products, one recurring action is: renaming.

Baidu's Ernie Bot was first renamed to Wenxiaoyan, and later renamed to Wenxin; Alibaba's Tongyi Qianwen was first simplified to Tongyi, and later renamed to Qianwen. From the products' public release to their second renaming, the two took 1029 days and 793 days respectively.

Changes in names are often accompanied by the company's rearrangement of product positioning and business boundaries.

About ten months after Minimax's Hailuo AI launched, its general assistant was renamed MiniMax, while the original "Hailuo AI" brand shifted to the video business. The name still exists, but the primary use it points to has changed.

About ten months after ByteDance's MarsCode programming assistant was released in China, it was renamed the TRAE plugin. With the launch of the TRAE standalone development environment, the original programming assistant was placed into a larger product system and became one of its usage modes.

Quark continued to revise its existing product. From the upgrade of the "Super Search Box" in July 2024 to the release of the "AI Super Box" in March 2025, 246 days passed. It attempts to expand the search entry point familiar to users into an AI application capable of handling more tasks.

Of course, some products did not get the chance to be "renamed." Tencent's QClaw opened public beta in March 2026, and by September it announced service shutdown arrangements (planned to stop service in December), only 188 days from public beta to the announcement of shutdown. A lifecycle of less than half a year reflects that the fault tolerance for product trial and error in new tracks is dropping sharply.

For companies, consolidation can concentrate investment and reduce duplicate construction; for users, frequent renaming and migration also mean re-identifying entry points and adapting to operating methods.

General assistants increasingly resemble a "hodgepodge of features"

As standalone products continue to differentiate and consolidate, more and more things can be done in general assistants.

From chatting to web search, file understanding, and then to drawing, voice, deep research, document generation, and code execution, product directions that were originally discussed separately are gradually being packed into the same assistant.

Public records of four samples show that by the end of 2024, Doubao, Qianwen, DeepSeek, and Kimi all had the three capabilities of chatting, web search, and file understanding. After that, research, office, and execution capabilities continued to enter some assistants. By October 2026, Doubao, Qianwen, and Kimi had all launched document generation and code or tool execution capabilities, and the product differences among general assistants were becoming smaller and smaller.

Standalone products need to reacquire users and cultivate usage habits. Existing assistants adding features can reuse accounts, entry points, and user relationships. For users, a complete task often spans multiple directions—first searching for materials, then reading files, and finally making a report or PPT. Being able to complete these steps continuously may be more attractive than having multiple tools that do not interconnect.

However, as more and more general assistants can search, write, and generate content, differentiation is no longer determined only by feature lists. Whether results are reliable, whether tasks can be completed continuously, how many times users need to repeatedly correct them, and whether service prices are reasonable will all affect which one users ultimately stay with.

The wave has surged for four years, and the ten major tracks are already crowded with products. QClaw's 188-day exit is not an isolated case, but the norm of this brutal competition. Old products are exiting one after another or moving toward consolidation, while the competition among companies over the next-generation super entry point has only just begun.

This article comes from the WeChat public account "Jingxiang Studio", author: Huang Yiting, published by 36Kr with authorization.