Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Most of us have encountered artificial intelligence in the form of the chatbot. You ask a question and it gives you an answer. The exchange is low stakes because even if the AI is wrong the consequences are usually limited.
Agentic AI breaks that model because it acts on its own. Give it a task and it will search, compare, decide and execute across digital systems on your behalf. The difference is that a chatbot recommends a flight while an agentic AI pays for it and sends you the confirmation.
Until now, mainstream use of these agents has been seen as some way off. But the speed at which they are being deployed in China is bringing that timeline forward much faster.
China’s technology giants Alibaba, Tencent, Baidu and a growing number of start-ups are trying to make agentic systems easier to build and integrate into daily life. The open-source AI agent OpenClaw developed by Austrian Peter Steinberger is driving a wave of experimentation among Chinese users as they begin delegating everyday digital tasks to AI. Local governments have rolled out subsidies and pilot programmes to accelerate adoption. Local tech groups have allowed users to run OpenClaw on their cloud systems while Baidu has integrated it directly into its main search app, bringing it to more than 700mn monthly active users.
Alibaba has also launched an AI platform for enterprises targeting automation this week, underscoring how quickly competition in China’s AI agent market is escalating. Its new Wukong platform, designed to co-ordinate multiple AI agents across business tasks, signals a shift from experimentation to deployment in enterprise workflows.
The advantage China has in developing agentic AI is that these agents are most valuable when they can act. This includes completing purchases, transferring money and coordinating services, all of which requires seamless end-to-end integration across payments, logistics, messaging and ecommerce apps. These are already consolidated within China’s super apps run by companies such as Alibaba and Tencent, including WeChat, which has about 1.4bn monthly active users.
For companies like Alibaba, this creates a self-reinforcing cycle across business units. More agents increase demand for cloud services and drive more activity on its marketplaces. If agents handle everything, users are less likely to leave the company’s platforms.
What will follow is a new economic model. Agentic AI could expand the monetisation of AI beyond subscriptions into continuous activity. Each task, transaction or workflow an agent performs requires computing power. Unlike traditional software, which is used intermittently, agents operate continuously, using resources as they plan and act. Instead of charging users for access, platforms can increasingly charge for completed transactions or executed workflows, in effect a form of metered labour.
But the same shift that makes agentic AI powerful also makes it far more fragile than the hype suggests. Early agents remain prone to misinterpretation and security flaws that allow malicious inputs to make an agent do the wrong thing.
More concerning is how easily they over-reach. Giving software access to payments, user accounts and enterprise systems can involve scanning files and reading messages in ways users do not fully control, with real-world consequences such as unauthorised payments or data leaks. Regulatory questions remain, including compliance in sectors such as finance and healthcare and who is accountable when agents make errors. There is also a risk that users place too much trust in systems that are not yet reliable.
Given the speed of recent rollouts, China will probably be both the testing ground and a leading indicator for agentic AI. In the US, the different parts needed to run AI systems are often controlled by separate companies. AI model developers, cloud providers and apps are separated as are payments, commerce and messaging services. A similar dynamic exists in Europe, where regulatory constraints can make integration harder. That fragmentation makes agentic AI harder to deploy at scale, as systems must navigate across multiple providers.
Until now, much of the conversation about who leads the AI race has focused on model capability: who scores highest on controlled benchmarks. The US still holds the lead in models. But once AI begins to act, benchmark scores matter less than the ability to get things done. By that standard, China may already have an edge.
