Chinese media are using the informal term “handcrafted economy” for individuals and tiny teams that build products with AI and open-source tools, while Suzhou has gone further by making AI-assisted one-person companies part of its economic policy. The city’s 2025-2028 plan targets 1,000 new AI OPC businesses, more than 10,000 people in its OPC ecosystem and over 50 municipal-level OPC communities by 2028. The model could widen access to entrepreneurship, but it does not remove dependence on the companies that supply AI models, cloud infrastructure and app distribution.
China Daily has used “handcrafted economy” as an English rendering of the Chinese label shoucuo. The word originally referred to completing difficult actions manually in video games. Chinese broadcaster CCTV described the term as an informal expression rather than a formal economic concept.
In its current use, the phrase refers to individuals or small teams combining AI, open-source software and readily available digital services to turn ideas into products without the resources of a large technology company.
That distinction matters. “Handcrafted economy” is a media label. Suzhou’s AI One Person Company programme is an official local government policy. Neither establishes that a new sector has already become economically large or that most solo AI businesses are commercially successful.
Tiny apps can still find large audiences
One example is Sileme, a Chinese app whose name roughly translates as “Are You Dead?” It asks people living alone to check in regularly and can notify an emergency contact after consecutive missed check-ins.
Reuters reported on January 14 that the app was second in Apple’s paid-app chart in China after reaching the top earlier that week. The operation behind it said it was independently run by a three-person team whose members were born after 1995.
The developers announced Demumu as the product’s global brand and said they would introduce an eight yuan payment scheme to help cover rising costs. A high chart position does not reveal how many people remained active, how much revenue the app produced or whether the attention lasted. It does show that a narrowly focused product can attract a national audience without a large development company behind it.
Another example is Cat Fill Light, which turns a phone screen into an adjustable light for selfies and photography.
CCTV profiled its creator, Chen Yunfei, who said he had not previously known how to code. According to the broadcaster, he used AI tools to create an initial version in just over an hour in 2024, then refined the idea after his girlfriend suggested using the screen as a photographic fill light. A paid version priced at one yuan later topped the paid-app ranking in an unnamed app store.
As of August 11, 2026, Google Play listed Cat Fill Light with more than 100,000 Android downloads. That figure covers Google Play only and should not be treated as a verified total across all platforms.
Neither app is technically comparable with a frontier AI model or a large enterprise software system. Their business relevance comes from their narrow scope. Each addresses a specific problem with a small product, which is precisely the kind of project that becomes more practical when development costs fall.
Suzhou is treating solo AI businesses as economic policy
Suzhou formalised its approach in December 2025 with an AI OPC development plan covering 2025 to 2028. OPC stands for One Person Company.
The city defines an AI OPC as a form of organisation in which an individual, supported by AI, can independently handle activities ranging from product design and development to production, market launch, user operations and customer service.
By 2028, Suzhou wants to cultivate 1,000 new OPC businesses, bring together more than 10,000 people in its OPC ecosystem, establish over 50 municipal-level OPC communities and place more than 100 tools on a public service platform.
The plan covers more than software applications. It identifies digital marketing, education, ecommerce, manufacturing, industrial design, healthcare, financial technology, connected vehicles, smart devices, data services and model deployment among the areas where OPCs could operate.
The support package shows the limits of the one-person label. Suzhou plans to offer access to computing capacity, models, datasets, low-cost offices, open-source services, financing connections and legal, tax and intellectual-property support.
A one-person company may have no employees, but it still depends on infrastructure, finance, regulation and distribution. Suzhou’s policy recognises that lower staffing requirements do not eliminate the need for an economic support system.
AI coding is widespread, but productivity is uneven
AI development tools have moved well beyond experimentation. In a company-run survey conducted in January 2026, JetBrains reported that 90% of more than 10,000 professional developers regularly used at least one AI tool for coding and development work. Some 74% had adopted specialist tools such as coding assistants, AI editors or agents.
For people without conventional programming backgrounds, these systems can translate natural-language instructions into code, explain unfamiliar concepts and help identify faults. This approach is often called vibe coding.
The phrase can make the work sound almost automatic. The evidence is more restrained. A July 2026 preprint reviewing 47 academic and professional sources described vibe coding as an iterative process of specifying a task, generating code, evaluating the result and revising it.
Twenty-one of the 47 sources reported short-term productivity or faster prototyping. Evidence was strongest for prototypes and user-interface work, while evidence on maintainability, long-term quality and use in production or safety-critical systems remained limited.
Individual experiments have also produced different results.
In a randomised trial involving 96 Google software engineers, people given access to three AI features completed a complex task in an average of 96 minutes, compared with 114 minutes for the group without those features. After accounting for other factors, the researchers estimated a 21% to 26% reduction in task time. However, the confidence intervals were wide and the preferred adjusted estimate did not meet the conventional 5% threshold for statistical significance.
A separate randomised study by METR reached the opposite result. Sixteen experienced open-source developers worked on 246 real issues in software repositories they knew well. When AI tools were allowed, they took 19% longer on average.
METR warned against applying that finding to most software work. Its participants were experienced developers modifying large, mature codebases with demanding quality standards. Building a simple prototype from scratch is a different task.
The trust problem remains. The 2025 Stack Overflow Developer Survey found that 84% of respondents were using or planning to use AI tools. However, 46% distrusted the accuracy of their output, compared with 33% who trusted it. Only 3% reported a high level of trust.
AI is therefore better understood as a tool whose economic effect depends on the user, the task and the required quality. It can lower the entry barrier for some projects without removing the need for testing, security checks, maintenance and technical judgement.
When code becomes cheaper, attention becomes scarcer
Lower production costs do not automatically create customers.
If more people can build applications, the supply of applications can rise. Discovery, reputation and distribution then become more important. A creator may spend less on programming but more time finding users, explaining the product, supporting customers and proving that the software is trustworthy.
Apple’s own rules show why differentiation matters. Its App Review Guidelines say developers should not submit apps that are indistinguishable from products already widely available. Apple warns that opportunistic variants damage discovery and reduce the quality of the store.
That policy predates the current wave of AI coding tools and does not prove that app stores are being flooded with AI-generated copies. It shows that Apple already treats low-effort duplication as a distribution problem, one that cheaper software production may intensify.
This is where people with industry knowledge may have an advantage. An accountant, teacher or hotel operator may understand a narrow operational problem better than a general software company. AI can reduce the technical distance between recognising that problem and testing a product that addresses it.
Even so, knowing the customer remains separate from writing the code. A useful application can still fail because nobody discovers it, users do not trust it, support is poor or the business cannot retain paying customers.
Large platforms may benefit from smaller producers
The handcrafted economy does not mean individuals are replacing the organisations that build and operate frontier AI systems.
The Stanford AI Index reported that industry produced more than 90% of notable frontier models in 2025. Training and operating the most capable systems still requires large amounts of computing capacity, data, energy and capital.
A solo developer may compete with a larger software company for a particular customer while still relying on an AI model provider, a cloud company, an operating system, an app store and a payment processor. Those suppliers can set prices, change technical rules and determine whether a product is admitted to a marketplace.
In mainland China, Apple reduced its App Store commissions in March 2026. The standard rate for paid apps and in-app purchases fell from 30% to 25%, while the rate for qualifying Small Business Program transactions fell from 15% to 12%.
The lower rates leave more revenue with developers, but Apple still controls review, ranking and access to its store. Google’s rules differ by market and transaction type. Under its user-choice billing programme, service fees continue to apply even when an eligible developer offers another billing system.
Creation may become more decentralised while infrastructure and distribution remain in the hands of much larger suppliers. Millions of small developers could create more competition among products and, at the same time, become more customers for AI, cloud and marketplace providers.
The model challenges the assumption that a software producer must itself be large. The underlying platforms may retain much of their influence.
Could this become a new form of work?
AI-assisted microbusinesses could broaden self-employment by making it cheaper to test specialist products. An individual might maintain several small applications or digital services rather than building one conventional startup around a single large product.
This model could support markets that are too narrow for a large software company but large enough to sustain a small operator. It may also favour people whose main advantage is practical knowledge of a profession, industry or local problem rather than formal software training.
However, the available examples do not show how many solo AI businesses produce a durable income. A chart-topping app and a government target demonstrate possibility and policy interest. They do not establish typical revenue, survival rates or earnings.
Nor does cheaper experimentation guarantee better products. As the cost of launching software falls, the cost of producing unsuccessful software falls as well. More attempts may create more useful niche tools, but they may also create a much larger number of products with few users.
Suzhou’s 2028 targets will provide one test of the model. The number of registered OPC businesses will be only part of the answer. Their survival, revenue, customer retention and the income earned by their owners will show whether AI has created a durable form of one-person enterprise or mainly made it easier to launch another app.
