AI代理的竞争与早期的AI大模型之争截然不同，它不仅需要更强大的模型，还需要信任机制、身份认证、支付系统以及对真实世界服务的接入。微信生态系统已具备这些基础设施，这是竞争对手难以从零搭建的优势。然而，风险同样巨大——一个在中国主导超级应用内出错的AI代理可能会大规模地破坏用户的信任，因此腾讯采取了谨慎、小范围的测试策略。
腾讯的成功并非板上钉钉。阿里巴巴旗下蚂蚁集团正在支付宝中测试类似的AI代理功能，字节跳动的豆包在下载量上也远超腾讯的元宝。此外，为十亿级用户规模运行AI代理所需的巨大算力也是一大挑战。尽管如此，几乎没有哪家科技公司比腾讯更有机会让AI代理在如此大的规模上真正落地——如果成功，中国的杀手级AI应用无需从零开始构建，因为它已经存在于14亿人的口袋里。
Tencent Holdings confirmed this week that it is beta-testing a native AI assistant called Xiaowei inside WeChat, the super app used by over a billion people that serves as digital infrastructure for daily life in China. Long seen as a laggard in the domestic AI race compared to rivals like Alibaba and ByteDance, Tencent is now revealing a strategic hand it has been holding back. The news sent investor excitement surging, with earlier reports of the WeChat agent driving the stock to its highest intra-day jump since 2022.
The race to build AI agents differs fundamentally from the initial scramble to develop large language models. Agents require trust, identity verification, payment integration, and access to real-world services—infrastructure that WeChat's ecosystem already possesses and that competitors would struggle to replicate from scratch. However, the stakes are equally high: a malfunctioning agent inside China's dominant super app could shatter user trust on a massive scale, which is why Tencent has opted for a cautious, limited rollout during the beta phase.
Tencent's success is far from guaranteed. Ant Group is testing a similar agent within Alipay, and ByteDance's Doubao app has garnered more than double the downloads of Tencent's Yuanbao this year. The enormous compute power required to run agents at WeChat's billion-user scale poses another significant hurdle. Nevertheless, few companies are better positioned to make agentic AI truly useful—or unavoidable—at such scale. If Tencent pulls it off, China's killer AI app will not need to be built from the ground up; it will already be sitting in 1.4 billion pockets.