In the AI era, the slower you learn, the less outdated material you have to learn. Author |Li Na
Editor |Cen Feng
Each generation has its own eggs to claim. (Leiphone(WeChat Official Account: Leiphone))Previously, people lined up outside supermarkets to collect free eggs; now, they camp out in AI communities for invitation codes, scramble for beta-testing slots, and accumulate free credits. Every time a new Agent is released, people rush to find a way to try it, figure out how to claim points and squeeze out more tokens, and fear missing the next epoch-defining product that changes the way we work.Yet some of the AI products in this lineup are already preparing to bow out before the rest of us have even gotten used to them.
Tencent's QClaw, which only opened for beta testing in March, announced at the end of September that it would cease operations by the end of the year. From scrambling for invitation codes to claiming points and moving to another platform, the whole thing lasted no more than 199 days.There have been several similar changes: In October, Yuanqi, another platform for creating agents, also posted a service shutdown announcement. Alibaba, meanwhile, reintegrated product lines such as Wukong, QoderWork, and MuleRun into Qwen Office.Science fiction writer William Gibson once said:“The future is already here; it is just not evenly distributed.” In today’s AI application market, another kind of time lag is beginning to show: before the craze for a new product has reached all users, the first early adopters have already received migration notices.
science fiction author William GibsonThis uncertainty is not limited to new Agents that have only been out for a few months.On October 12, the once-phenomenally popular AI companion platform Character.AI announced that it had filed for Chapter 11 bankruptcy protection in the United States, seeking a court-supervised sale of the company’s assets. The product, which still has about 12 million active users, is currently operating normally, and users can continue to log in and chat with their familiar AI characters. But if no acquisition plan capable of keeping the platform running is ultimately found, Character.AI will still face closure.
On one side are the endlessly emerging new models and new Agents; on the other side are old products that, after barely a short time of use, have begun to be adjusted, consolidated, or even exited. In the related discussion forum, one netizen left a rather interesting comment:“In the AI era, as long as you learn slowly, you get to learn a lot less outdated stuff.”
01
The big tech giants’ Agent race is winding downTAG142TAG144TAG146So far this year, the big tech giants have released Agents at a fairly rapid pace, but some products have not even been validated for half a year before a new round of internal adjustments began.We reviewed the 16 representative AI products and new product formats publicly launched this year by Tencent, Alibaba, ByteDance, and Baidu. As of October 19, one product had announced that it would discontinue service in the future, and five others had undergone product line consolidation, a halt to new purchases, or organizational restructuring.The fastest to change was ByteDance’s TRAE Work, which went from being upgraded into a new product format in June to having its team merged into the Doubao system in August, a gap of only 76 daysTAG165@@; and Wukong, launched by Alibaba in March, was only 107 daysTAG171TAG173 away from the start of integration across its three product lines.
Alibaba is a typical case. At the beginning of the year, QoderWork, Wukong, and MuleRun were each being driven by different teams, focusing respectively on desktop execution, enterprise collaboration, and cloud-based tasks. As their capabilities expanded, their office scenarios gradually began to overlap. Caixin previously cited sources inside Alibaba who said that the early multi-team exploration had its justification, but at this stage there was no longer any need to keep “reinventing the wheelTAG184TAG187@@.”TAG190@@In July, the three product lines began to be integrated; in August, Qianwen Office began bringing the previously scattered capabilities together under a single entry point.Tencent is also consolidating its product lines. QClaw and WorkBuddy, launched almost simultaneously in March, initially focused respectively on remotely controlling a computer via WeChat and on all-scenario AI office work.In July, businesses related to QClaw and some teams were reassigned to the department to which WorkBuddy belongs; in September, Tencent announced that QClaw would cease operations by the end of the year and that a channel for migrating to WorkBuddy would be provided. ByteDance, meanwhile, integrated the teams related to TRAE and Coze into the Doubao ecosystem in August. Its existing coding product line continued to develop, while office Agent capabilities were consolidated into Doubao Workspace. It is easy to understand the intense competition at the beginning of the year. After the OpenClaw boom, no one has a clear answer as to whether desktop Agents, cloud-based tasks, or enterprise office platforms will become the next-generation entry point. Rather than betting on a single path, multiple teams are advancing simultaneously, seizing the market first and then seeing which product emerges as a winner. But as the first wave of products is rolled out,computingcostshave become a major problem, keeping multiple Agent products without clear business models running in parallel is an enormous waste of GPU resources. Consolidating scattered entry points is essentially a way to pool computing power, drive up daily active users, and present a more complete business story to enterprise clients. Besides costs such as R&D, maintenance, and model calls, iterations in foundation models are also changing the space in which standalone Agents can survive. Kin, SpaceX’s China ambassador and an AI entrepreneur, said in an interview with AI Tech Review that the pace at which Agents are replaced has always been fast and did not suddenly accelerate this year. As foundation models such as GPT and Claude continue to be upgraded, the original capabilities of Agents that lack unique advantages may soon be eclipsed by new models or all-in-one products. However, product culling does not mean that users’ enthusiasm for trying new things is fading. Kin mentioned that in the dozens of AI-related WeChat groups he belongs to, as long as a new product has something noteworthy, everyone is still willing to try it immediately. The real question arises after that initial trial: whether an Agent can provide enough long-term value to keep users coming back to it, rather than have them try it a few times and then put it aside. 
02
A product has users—why can’t it keep them?
For companies such as Tencent and Alibaba, discontinuing a standalone product does not mean abandoning the underlying 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 Workspace. Major tech companies have enough businesses and teams to reallocate resources and continue searching for the next-generation entry point for AI. But independent startups may not have such a fallback. 
Character.AI was once one of the most closely watched companies in the AI companion sector. In 2021, former Google researchers Noam Shazeer and Daniel De Freitas founded the company, which quickly rose to fame with a product format that allowed users to freely create characters and have long-term conversations with AI. By 2024, it had reached around 20 million monthly active users, and its valuation had once reached $1 billion. This former star AI unicorn was not short on users. According to bankruptcy documents cited by Reuters, it still has around 12 million active users, but it is losing as much as $8 million to $9 million per month. In 2025, the company was still spending around $10 million per month on computing power, while monthly subscription revenue peaked at only around $2.8 million. Even with advertising, revenue consistently struggled to cover its high operating costs. 
In addition to computing expenses, Character.AI also faces multiple pressures, including the departure of its founding team, technological competition, and legal lawsuits.The highly engaged user base that once propelled it to rapid popularity ultimately failed to support a sustainable business. Today, the company can only look to asset sales to find the platform a new way forward.By contrast, in the Agent race among major technology companies, even when a particular product exits, its underlying technology and users may still have opportunities to move into a new product ecosystem; Character.AI, however, faces the question of whether the platform as a whole can continue operating.Behind these two situations is an easily overlooked fact: even if an AI product still has users, that does not mean the company necessarily has a reason, or the ability, to continue operating it. For QClaw users, there is at least WorkBuddy to migrate to; Character.AI users must wait for the final outcome of the bankruptcy proceedings to learn whether the platform they know will continue to exist.
03
When a Product Exits, What Can Users Take With Them?
Once a product announces its exit, the clock begins ticking for its users.On September 24, QClaw announced that it would cease operations, with the backup and migration window remaining open until March of the following year. Tencent Yuanqi will stop creating new agents and distributing API Keys on October 29, officially shut down its services in November, and keep its data export window open until January of the following year.Data can be exported, but the original workflow may not be fully migrated.According to the ADP migration guidelines published by Tencent Yuanqi, the knowledge base, workflows, and Widget in the new version of Yuanqi can be exported and imported directly, but agent settings and custom tools still have to be configured manually; workflows from older versions must be rebuilt according to their original logic. Because the two platforms do not support exactly the same models and plugins, the results after migration must also be revalidated.
Some products have also set retention periods for data.In July of this year, Doubao announced that it would discontinue the feature allowing users to build their own agents, giving users a transition period in which to view and save relevant information and historical conversations. According to the announcement, after October 15, users will no longer be able to view or recover that data through Doubao.For users who have used Agents over the long term, switching products does not necessarily mean starting from scratch.AI enthusiast Likka began paying to use Cursor in April of this year. What initially attracted him was the ability to use different models within the same development environment and to view and modify code at any time. But as the product and its services changed, he began looking for alternative tools.Project files, AGENTS.md, and Skills can all be packaged and taken along, but what truly makes switching difficult is his usage habits. He has tried products such as Codex and Claude Code, but still relies primarily on Cursor. “When you switch to a new platform, you will feel unfamiliar with it. That may be a kind of sunk cost, too.”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 it had been running for a while, a succession of problems involving interfaces, configurations, and maintenance arose, and he ultimately gave up.Another interviewee, Kin, is more willing to proactively bear the costs of migration. He is accustomed to backing up project context, Skills, and other materials to GitHub and cloud storage, and preparing alternatives for the tools he uses most. Some of the work he previously completed in Codex and Claude Code has now been moved to Grok Bot, allowing him to continue working through his phone and iPad while on the go.In Kin’s view, as long as a new product delivers efficiency gains as tangible as those of Grok Bot, it is worth trying.The experiences of these two illustrate the different choices Agent users face when products are replaced: some keep moving to better tools, while others remain on familiar platforms. But either way, the cost of migration is not merely a matter of exporting data.
04
Ending
From OpenClaw setting off a craze early in the year, to major tech companies gradually consolidating their product portfolios, and even to Character.AI—with tens of millions of users—reaching the point of bankruptcy and restructuring, the fate of AI products seems harder to predict than people imagined.Vendors are busy searching for the next entry point, while users chase new products and have to consider how to move the workflows they have accumulated. But at least according to the respondents, these changes have not dampened their enthusiasm for new technology.Likka likens the current Agent competition to a marathon: some are running this way, others are running that way, and no one knows exactly where the door to the next stage is. But you cannot stop running simply because you don't know.New products are still emerging one after another. No one knows where the next door is, but clearly everyone still wants to keep searching.
All aboard! Let us take you through the highlights of the world's top AI conferences.
Get exclusive access to:
Expert presentation slides
Full conference reports
Trending paper analyses
Interviews with rising academic stars
Scan the QR code above
or click
“Read the original article”to follow the dedicated section.
An original article from LeiFeng.com. Unauthorized reposting is prohibited. For details, seethe Reposting Guidelines.
