What happened
36Kr reports that WeChat is conducting a small-scale gray test of Xiaowei AI Social. With authorization from both participants, their Xiaowei assistants can communicate in an independent session to coordinate times, synchronize information, or hand off tasks, then return a proposed result to users for final decisions. The communication does not appear in the participants’ ordinary WeChat chat windows, according to the report.
According to 36Kr, Xiaowei AI Social creates an authorized, separate session between the Xiaowei assistants of two users. The assistants can handle preliminary coordination, such as arranging a meal time, before showing the result to the people involved for confirmation. The report says both users must qualify for Xiaowei and authorize the exchange.
Newrank, whose article was republished by 36Kr, says its editorial team tested the feature with a meal appointment and received a confirmation card after the two assistants communicated. This is a report-specific practical test, not an independently reproduced result. The source does not provide technical performance data or a full description of failure handling.
The same WeChat update reportedly adds an “AI Processing” entry for images in chats, including information extraction, beautification, and image modification on the current page. Other actions, such as remembering or interpreting images and generating emojis, reportedly open Xiaowei for further processing. A search-page control was also renamed “Hold to ask Xiaowei,” according to 36Kr.
Why it matters
The reported test shifts Xiaowei from a personal assistant toward an intermediary that can coordinate between people. That could reduce the friction of routine scheduling, but it also introduces questions about consent, context, privacy, accountability, and when an AI should stop negotiating and return control to a person. The feature is not broadly available, and the reported behavior has not been independently verified.
The practical change is not merely that WeChat can answer questions. The reported design allows AI systems to represent users during an authorized negotiation, creating a human-AI-AI-human workflow. For routine, low-emotion tasks such as scheduling, that may save time and reduce repetitive messages.
The same design could create risks if an assistant misunderstands availability, social expectations, sensitive information, or the intended tone of a message. The source says the assistants do not send ordinary WeChat messages, but it does not independently establish what data are exchanged, how long records remain available, or how users can inspect or correct the process.
The update also places AI closer to existing search and chat-image workflows. That may make AI features easier to use in context, while increasing the importance of clear permissions and boundaries around image analysis and retained user information.
What to watch next
Watch whether WeChat expands access beyond the small test, clarifies how authorization and records are handled, and limits the assistant to low-risk coordination. Pricing, broader availability, supported languages, retention policies, and performance outside the reported meal-planning example are unknown.
The immediate unknown is rollout. 36Kr describes Xiaowei AI Social as a small-scale gray test, and says the feature works only when both parties have access to Xiaowei. The source does not establish general availability, a launch date, pricing, or whether access is limited by region or account.
Further reporting should verify the feature directly and examine authorization prompts, the independent-session record, user controls, data retention, and behavior when the assistants cannot reach agreement. It should also test whether the system can distinguish harmless scheduling from decisions requiring human judgment.
The reported audio entry for official-account content is a separate product change rather than the central AI development. Its traffic impact and recommendation quality remain unknown, and the source offers no independent measurement.