Commentary: Can WeChat become China’s killer AI app?

Commentary: Can WeChat become China’s killer AI app? If Tencent can manage it, China’s killer AI app won’t need to be built from scratch, says Catherine Thorbecke for Bloomberg Opinion. TOKYO: One of the biggest problems for China’s artificial intelligence sector is that no firm has created a killer AI app. Tech giants and startups have flooded the market and created a lot of excitement, but the competition has made it hard for any of these releases to differentiate themselves. But what if it has been hiding in plain sight this whole time? Tencent confirmed this week that it’s testing a native “AI assistant” inside WeChat, the super app used by more than a billion people that’s also a kind of digital infrastructure for daily life in the world’s second-largest economy. The Shenzhen-based firm has long been cast as a laggard in the domestic AI scramble compared to more aggressive peers like Alibaba and ByteDance. It’s now revealing the hand it’s been saving this whole time. WeChat, also known as Weixin, represents the kind of moat that competitors can only dream of. And if done right, it could be the ideal ecosystem for launching agents - the next phase of the AI adoption race marked by systems that go beyond just chatbots and take action on a user’s behalf. THE RACE TO DEVELOP AI AGENTS Currently in beta testing and available to a small group, Tencent said Monday (Jun 22) that users can interact with the new assistant, dubbed Xiaowei, via text or voice within the app. It may be the first step in Tencent’s AI resurgence. Investors understand the stakes. Reports earlier this month that Tencent was working on a WeChat agent sent shares soaring in their highest intra-day jump since 2022. In March, The Information reported that the company had been building a new AI agent in its super app as part of a “high-priority, top-secret initiative” since late last year. The hush-hush approach makes sense for a company that can’t afford any missteps here. WeChat isn’t just another social platform but a place where people do everything from paying bills and running businesses. The race to develop agents is different from the initial AI scramble. It’s not just about making models bigger and better, but requires trust, identity, payments and access to real-world services to be useful. The WeChat ecosystem already has much of that plumbing for building agentic infrastructure that will be harder for competitors to create from scratch. At the same time, the cost of failure is significantly higher. The outputs of an unreliable chatbot can be ignored, but a rogue agent that makes a mistake inside China’s dominant super app risks breaking institutional trust at a massive scale. It calls for a cautious, limited rollout. OPPORTUNITY FOR SUCCESS Chinese consumers have shown a willingness to experiment with nascent agentic tools. An OpenClaw fervour enamoured the country earlier this year, propelled in part by a fear of falling behind. Tencent moved quickly, offering services to help users install the open-source software and launching agentic tools such as QClaw and WorkBuddy. But it’s one thing for people to test out an agent and another for ordinary users to form a habit of trusting them every day inside WeChat. Every signal from the Xiaowei beta testing will matter, including what users are willing to ask it to do, where they hesitate, and whether it ultimately makes WeChat feel more useful or intrusive. Its success is far from inevitable. Alibaba’s Ant Group is similarly testing an agent within super-app rival Alipay that can be used to book cars or order coffee. Competitors’ AI apps have resonated with consumers a lot more than Tencent’s offerings. ByteDance’s Doubao has more than double the downloads year-to-date of Tencent’s Yuanbao, according to Sensor Tower data. Downloads for Doubao plunged in May after ByteDance introduced paid premium tiers, suggesting that consumers still don’t want to pay for AI features - this could present a problem given costs for agentic software can balloon quickly. Another challenge is compute. Running agents inside an app used by a billion people would require immense processing power. It’s not yet clear that Tencent can access such capacity. Still, few tech companies have a better shot at making agentic AI actually useful, or even just unavoidable, at this scale. If Tencent can manage it, China’s killer AI app won’t need to be built from scratch. It will already be sitting in 1.4 billion pockets.