Doubao ve Qwen Shut Down the AI Agents They Built for Two Years

ByteDance's Doubao ve Alibaba's Qwen platforms announced their decision to remove artificial intelligence agent (AI agent) functionalities on the same day and with the same deadline. According to TMTPost, this decision aligns perfectly with China's new anthropomorphic artificial intelligence interactive services regulation, which went into effect on 15 Temmuz. This coordinated move by China's two largest artificial intelligence assistant platforms brings an end to the two-year industry hype cycle surrounding consumer-focused artificial intelligence agents. The regulation includes strict requirements such as real-name verification, content filtering, and minor protection measures. By choosing to remove these features entirely rather than making them compliant with the regulation, the companies clearly revealed their internal priority calculations.

The artificial intelligence agent trend was heavily promoted across the industry as a new era of generational human-computer interaction paradigms. While Doubao created its own agent plaza, Qwen offered users character customization opportunities; Tencent's Yuanbao platform launched artificial intelligence applications, and even NetEase developed emotional companion applications. User engagement data, such as time spent on platforms and rapidly increasing character creation metrics, looked highly impressive. Users could create customized artificial intelligence personalities in these marketplaces, assign them tasks, and allow the artificial intelligence to operate autonomously. However, behind this bright picture lay two fundamentally different product categories that the industry had conflated for two years. Unfortunately, this consumer-side growth was not based on a sustainable economic model in the long run.

Behind the positive engagement figures, there was an unsustainable economic balance straining the companies. Internal estimates indicated that agent features generated less than 10 cents in direct revenue for every yuan of computing cost. The most popular use case, namely anthropomorphic role-playing and casual chatting, consumed massive amounts of inference computing power while contributing almost nothing in terms of commercial revenue. Every conversation users had with their artificial intelligence companions during late hours directly reflected on the platforms' GPU bills. Companies' attempts to introduce advertisements to offset these costs were met with immediate and severe backlash from users. This structural economic issue went beyond a mere technical barrier, exposing the fundamental vulnerability of consumer-focused anthropomorphic interaction.

The regulatory deadline provided a convenient exit door for the companies, and Doubao executed a particularly strategic maneuver. Instead of completely eliminating the agent features, Doubao migrated them to CatBox, a separate ByteDance product with its own content moderation system and user agreement. This move effectively disables the agent functionality in the flagship application, which cannot afford to take compliance risks. The features continue to live on by being transferred to a more background product where they can operate under different rules. However, this strategic retreat still demonstrates that the consumer-focused artificial intelligence agents market is experiencing a general reality check. Qwen similarly rearranged its priorities, deciding that these features did not rank high enough to be worth investing in.

The shelving of consumer agent features in this manner signifies an industry-wide awakening and pivot. C-end (consumer-side) anthropomorphic character interaction faces severe structural challenges such as high computing costs, poor monetization potential, and increasing regulatory pressures. In contrast, B-end (enterprise-side) corporate artificial intelligence agents are gaining speed and momentum. In Haziran, Qwen opened its corporate API access, which was piloted by brands such as Luckin Coffee, KFC, and Çin Doğu Havayolları. Task-oriented and tool-based agents, where return on investment can be measured and businesses are willing to pay, represent the true direction of growth. The industry is now moving forward by clearly separating these two fundamentally different product categories that it had conflated for two years.