AI Product Knockout: Big Tech Consolidates Agent Product Lines, Character.AI Files for Bankruptcy Reorganization

English text is being prepared.

English text is being prepared.

01

The Agent horse race among the big tech companies is beginning to wind down. This year, the big tech companies have been rolling out Agents at a fairly rapid pace, but before some products had even undergone half a year of validation, a new round of internal adjustments had already begun. We reviewed 16 representative AI products and new product forms publicly launched this year by Tencent, Alibaba, ByteDance, and Baidu. As of October 19, one of them had announced a future service shutdown, and another five had undergone product line integration, suspension of new purchases, or organizational adjustments. The fastest to change was ByteDance's TRAE Work: from its upgrade to a new product form in June to its team being merged into the Doubao system in August, only 76 days passed; Alibaba's Wukong, launched in March, was only 107 days away from the start of integration across three product lines. Alibaba is a typical case. At the beginning of the year, QoderWork, Wukong, and MuleRun were each advanced by different teams, focusing respectively on desktop execution, enterprise collaboration, and cloud tasks. As their capabilities expanded, their office scenarios gradually overlapped. Caixin previously cited sources inside Alibaba as saying that early multi-team exploration was necessary, but by this stage there was no longer any need to keep "reinventing the wheel." In July, the three product lines began to integrate; in August, Qwen Office was consolidating the previously scattered capabilities into a single entry point. Tencent is also consolidating its product lines. QClaw and WorkBuddy, launched almost simultaneously in March, initially focused respectively on remote computer operation via WeChat and all-scenario AI office work. In July, the QClaw-related business and some of its team were reassigned to the department to which WorkBuddy belongs; in September, Tencent announced that QClaw would cease operations at the end of the year and provided a channel for migration to WorkBuddy. ByteDance, for its part, integrated the TRAE and Coze-related teams into the Doubao system in August; the original programming product line continued to develop, and office Agent capabilities were consolidated into Doubao Work. The intense horse race at the start of the year is not hard to understand. After the OpenClaw boom, no one had a definite answer as to whether desktop Agents, cloud tasks, or enterprise office platforms would become the next-generation entry point. Rather than betting on a single path, it was better to have multiple teams push forward simultaneously, grab market share first, and then see which product could break out. But as the first round of products landed, computing power costs became a major problem; keeping multiple Agent products without clear business models running in parallel is a huge waste of GPU resources. Unifying scattered entry points is essentially about concentrating computing power, driving up daily active users, and telling a more complete business story to B-end customers. In addition to R&D and maintenance, model invocation, and other costs, the iteration of underlying models is also changing the living space of standalone Agents. Kin, SpaceX China ambassador and also an AI entrepreneur, said in an interview with AI Technology Review that the turnover of Agents has always been fast; it did not suddenly accelerate this year. As foundation models such as GPT and Claude continue to upgrade, some Agents lacking unique advantages may quickly see their original functions covered by new models or comprehensive products. However, product elimination does not mean users' enthusiasm for trying new things is fading. Kin mentioned that in the dozens of AI WeChat communities he is in, as long as a new product has highlights, people are still willing to try it at the first opportunity. The real question comes after trying it: whether an Agent can provide enough long-term value to keep users coming back, rather than being set aside after a few uses.

02

It has users, so why can't it retain a product? For companies like Tencent and Alibaba, ending an independent product does not mean abandoning the original business. Some of QClaw's capabilities can be taken over by WorkBuddy, and products such as Wukong and QoderWork can also be integrated into Qwen Office. Big tech companies have enough businesses and teams to reallocate resources and continue searching for the next-generation AI entry point. But independent startups may not have such a fallback. Character.AI was once one of the most closely watched companies in the AI companionship track. In 2021, two former Google researchers, Noam Shazeer and Daniel De Freitas, founded the company, and it quickly became popular with a product form that allowed users to freely create characters and hold long-term conversations with AI. By 2024, its monthly active users had reached about 20 million, and its valuation once reached $1 billion. This former star AI unicorn did not lack users. According to bankruptcy filings cited by Reuters, it still has about 12 million active users, but loses as much as $8 million to $9 million per month. In 2025, the company's monthly computing power spending is still about $10 million, while the peak monthly revenue of its subscription business is only about $2.8 million. Even with advertising added, revenue has always struggled to support high operating costs. In addition to computing power expenses, Character.AI also faces multiple pressures such as the departure of its founding team, technological competition, and legal lawsuits. The highly sticky user base that once made it quickly popular ultimately failed to support a sustainable business. Now, the company can only pin its hopes on an asset sale to find a new way out for the platform. By contrast, in the Agent horse race inside big tech companies, even if a product exits, the original technology and users still have a chance to enter a new product system; what Character.AI faces is whether the entire platform can continue to exist. Behind the two situations is an easily overlooked fact: even if an AI product still has users, it does not mean the company necessarily has a reason, or the ability, to keep operating it. For QClaw users, at least there is WorkBuddy to migrate to; Character.AI users need to wait for the final outcome of the bankruptcy proceedings before they know whether the platform they are familiar with will continue to exist.

03

When a product exits, what can users take away? After a product announces its exit, the time left for users has already begun counting down. On September 24, QClaw announced it would cease operations, with a backup and migration window kept open until March of the following year. Tencent Yuanqi will stop creating new agents and distributing API Keys on October 29, officially shut down in November, and extend the data export deadline to January of the following year. Data can be exported, but the original workflows may not be fully migratable. The ADP migration guide published by Tencent Yuanqi shows that the new version of Yuanqi's knowledge base, workflows, and Widget support direct import after export, but agent settings and custom tools still need to be configured manually; old-version workflows need to be rebuilt according to the original logic. Because the models and plugins supported by the two platforms are not exactly the same, the effect after migration also needs to be re-verified. Some products have also set data retention deadlines. In July this year, Doubao announced the shutdown of the user-built agent feature, allowing users to view and save related information and historical conversations during the transition period. According to the announcement, after October 15, users will no longer be able to view or recover these data through Doubao. For users who have long used Agents, product replacement does not necessarily mean starting over from scratch. AI enthusiast Likka began paying for Cursor in April this year. What initially attracted him was that Cursor allows different models to be used in the same development environment and makes it convenient to view and modify code at any time. But as the product and services changed, he began trying to find alternative tools. Project files, AGENTS.md, and Skills can all be packaged and taken away; what really makes switching difficult for him is usage habits. He has tried products such as Codex and Claude Code, but currently still mainly uses Cursor. "When you switch to a new platform, you will feel uncomfortable; this may also be a kind of sunk cost." He even tried to build a local relay system that did not depend on a single vendor, hoping to avoid losing his accumulated work because of a platform shutdown or account restrictions. But after running it for a while, various interface, configuration, and maintenance problems followed one after another, and he eventually gave up. Another interviewee, Kin, is more willing to actively bear the migration cost. He is used to backing up project context, Skills, and other materials to GitHub and cloud storage, and preparing alternatives for commonly used tools. Some of the work previously done in Codex and Claude Code has now been moved by him to Grok Bot, and he can continue processing it through his phone and iPad when out and about. In Kin's view, as long as a new product can bring sufficiently obvious efficiency improvements like Grok Bot, it is worth trying. The experiences of the two present the different choices Agent users make when facing product replacement: some keep migrating to better tools, while others remain on familiar platforms. But either way, the migration cost is not just a set of data that can be exported.

04

From the surge set off by OpenClaw at the beginning of the year to big tech companies gradually integrating product lines, and even Character.AI, which has tens of millions of users, reaching the stage of bankruptcy reorganization, the fate of AI products seems harder to predict than people imagine. Vendors are busy looking for the next entry point, while users chase new products and also have to consider how to move the workflows they have accumulated. But at least in the eyes of the interviewees, these changes have not worn down their enthusiasm for new technology. Likka compares the current Agent competition to a long-distance race: some run this way, some run that way, and no one knows where the door to the next stage is, but one cannot simply stop running because one does not know. New products are still constantly emerging. No one knows where the next door is, but everyone clearly still wants to keep looking.