A growing group of Chinese workers is using generative AI not just to complete tasks but to build simple tools of their own. An August 23 report in the South China Morning Post described non-programmers who have created agents and mini-programs after concluding that AI literacy were becoming a workplace expectation.
The story is not about a mass survey of China’s workforce. It is a report based on profiles collected by mainland outlet Jizhou Studio. But its examples capture a wider change in how ordinary users view AI. Instead of waiting for a software company to make a product for them, some people are using a model to sketch the logic, language, and structure of a tool that addresses a specific personal problem.
AI Literacy Anxiety Is Pushing Non-Programmers Toward Tool Building
SCMP says training courses and income-oriented tutorials now circulate widely online as AI competence becomes more important in many workplaces. That creates a familiar pressure: people who do not identify as programmers fear that new tools will become standard before they learn how to use them.
The response described in the article is active rather than passive. Generative AI and agents can help a user describe an application in ordinary language, break a problem into steps, and produce code or a prototype. That does not make software development effortless, but it can reduce the barrier for someone with a clear use case but no recent coding experience.
SCMP profiles Congcong, a 46-year-old former product manager, who used ChatGPT to make an English-practice assistant for her household. She later built a bot to monitor the child’s homework and breaks. These are narrow, personal tools, not venture-backed products. Their importance lies in the fact that a user turned a household need into an application without approaching a conventional software team.
The shift also helps explain why the public conversation around AI has become more practical. EastFrontier’s recent report on China’s AI podcast audience found that listeners were increasingly focused on people, work, and collaborative AI rather than only on model capabilities. Tool building is the next step beyond that discussion: users are testing how AI can change their own routines.
Personal Agents Make Everyday Problems More Programmable
The examples in SCMP are deliberately ordinary. An English-learning assistant and a homework-management bot do not require a frontier research breakthrough. They require a user to define a task, decide what information matters, and check whether the output is useful. That is precisely why they are significant. AI is making a portion of application design accessible to people whose previous role was simply to use software.
Jizhou Studio’s profiles, as described by SCMP, include people returning to ideas that would previously have required a developer. The newspaper says generative tools help users reason about requirements and coding even when they lack formal programming training. The model becomes an intermediary between an intention and a basic working product.
That intermediary is not infallible. A person making a tool with AI still has to identify errors, test whether the system does what it says, and decide what data should be included. The lower cost of prototyping can also produce many weak tools. But the ability to try is itself a change from a period when a non-programmer’s idea often stopped at a spreadsheet, a note, or a request to an IT department.
The limits are already visible in larger AI-agent products. EastFrontier’s reporting on agentic smartphones at WAIC showed that agents can face obstacles when they try to operate across apps and services. A personal agent may be easier to create than a reliable assistant that can securely take action in a complex digital environment.
China’s New AI Makers Will Need Judgment as Much as Prompts
The SCMP report frames this behavior through a fear of falling behind. That anxiety can lead people to chase every new model release or tutorial without considering whether a tool genuinely helps them. The more durable outcome will come when users choose a concrete problem, set boundaries for the AI system, and evaluate the result with ordinary judgment.
For employers, the rise of user-built tools has two implications. It can make workers more productive and surface useful process improvements. It can also create security, privacy, and quality-control problems if staff enter sensitive information into public services or rely on an untested agent for an important decision.
The technology therefore changes the meaning of AI literacy. It no longer means only knowing what a large language model is or being able to write a prompt. It increasingly means understanding enough about a task to decide what should be automated, enough about data to recognize risks, and enough about an output to verify it.
The people profiled by SCMP are not replacing professional developers. They are demonstrating that the boundary between a software user and a tool maker is becoming more porous. China’s AI market will not be shaped only by the companies releasing models. It will also be shaped by the workers who decide that a chatbot is not enough and use it to build something specific for themselves.