Cover Story: How AI Is Reshaping China’s Programming Landscape
By Du Zhihang, Liu Peilin, Guan Cong and Han Wei
Artificial Intelligence (AI) is reshaping industries worldwide, and programmers who build and refine this technology find themselves at the forefront of its disruptive impact.
In a rapidly evolving programming landscape, AI is automating tasks, streamlining processes and boosting efficiency — allowing users to create entire applications simply by describing their needs, with no coding required.
This wave of innovation, led by global internet giants, is forcing some 28 million programmers to confront a challenge that could ultimately redefine their careers.
On Sept. 6, California-based startup Replit launched Replit Agent, an app generation platform that claims to “turn ideas into fully functional applications without any coding experience.” Amjad Masad, the CEO of Replit, demonstrated how users can simply type requests into a browser dialog box, allowing the AI to automatically write code, create the app and adjust features based on further instructions — all without the need to input a single line of code.
Anysphere, an OpenAI-backed startup, recently saw its AI-powered coding app, Cursor, gain popularity on social media for its ability to generate webpages from simple language prompts. On Sept. 23, Sam Altman, the CEO of OpenAI, said that AI will significantly affect the labor market in coming years, though he expects most job transformations to progress more slowly than anticipated.
OpenAI, which developed ChatGPT, said on Oct. 2 that its popular chatbot is used by more than 250 million people worldwide each week to enhance their work.
The next day ChatGPT introduced “canvas,” a new feature for code development that enables programmers to make precise edits and can translate code between various programming languages.
Coding is a core strength of large AI models. Besides ChatGPT, models such as Google’s Gemini, Anthropic’s Claude and Meta’s Llama can all generate code from natural language prompts.
Since ChatGPT’s surge in popularity, AI-generated code has gained traction among programmers. “Often, I don’t even notice the AI assisting me,” a Google algorithm engineer told Caixin. His most-used AI feature on Google’s code editing platform is code completion, which can suggest revisions based on peer comments.
Executives are eager to accelerate AI integration into workflows. In May, IBM surveyed 3,000 CEOs across more than 20 countries and 26 industries, revealing that 61% are driving generative AI adoption. The CEOs expect that within three years, 35% of employees will need training or retraining, a jump from 6% in 2021. Furthermore, 59% believe competitive advantage hinges on having the most advanced generative AI, and 62% say they must rewrite their business strategies to stay ahead.
Meanwhile, employees are increasingly aware of AI’s potential to replace specific roles. According to Zhang Dandan, deputy dean at Peking University’s National School of Development, the jobs most at risk are accountants, auditors, editors, sales engineers, computer programmers and tax professionals.
“With the rise of large AI models, demand for mid-skilled workers will drop sharply,” Zhang told Caixin. “As automation grows and labor advantages decline, these workers will either need to move up to high-skilled positions or settle into more basic, low-skilled roles.”
A survey by ADP, the U.S. payroll and human capital management company, of nearly 35,000 respondents across 18 countries found that in China, 22% said AI occasionally assists with tasks, while 47% believed AI could replace some or all of their job functions. Among Chinese respondents in IT and telecommunications, these numbers increased to 25% and 51%. In the United States, the figures stood at 24% and 33%, respectively.
Programmers are at the forefront of AI’s rapid progress, guiding its integration across practically every stage of software development — from initial design to front-end and back-end development, testing, deployment and maintenance.
As a result, the business processes and talent needs of internet companies are quietly shifting. A tech recruiter in Shenzhen told Caixin that app and mini-program development workflows are now so streamlined most companies can outsource, reducing in-house roles to front-end development and testing. “Overall, the programmer job market now faces an oversupply,” the recruiter said.
For internet giants, AI is rapidly advancing production processes and reshaping business models. Since 2023, Chinese tech leaders such as Baidu Inc., Alibaba Group Holding Ltd., Tencent Inc., Ant Group Co. Ltd., SenseTime Group Ltd., and ByteDance Ltd. have all introduced AI coding assistants, while AI-generated content (AIGC) is increasingly being integrated into advertising and marketing workflows.
“We can’t view the future through the lens of the mobile internet era,” Alibaba chairman Eddie Wu said at the company’s annual cloud business conference in September. AI is creating new value on the supply side, driving global productivity, “this value generation could surpass that of the mobile internet by dozens of times,” he said.
AI-powered programming
Nearly all Chinese internet companies are integrating AI tools across business and product operations to boost efficiency. Baidu, the first in China to launch a large language model, led the way by releasing its AI coding tool Comate in June 2023.
According to Zang Zhi, the senior engineer who led the development of Comate, 40% of Baidu’s programmers now regularly use AI for unit testing. In July, Baidu CEO Robin Li announced that AI was generating 30% of Baidu’s internal code, with nearly 90% of programmers using Comate. In a bold — and even provocative —prediction, Li said that only two programming languages, Chinese and English, will remain in the future, envisioning the eventual obsolescence of the programming profession.
Alibaba-backed Ant Group is also advancing AI integration. In September 2023, Ant launched CodeFuse, an AI coding assistant designed to support the entire lifecycle of internal software development. According to Li Jianguo, the head of CodeFuse, AI can improve efficiency across the whole business chain of product development.
CodeFuse’s “Graph-to-Code” feature can convert design images directly into code. With just an app screenshot, developers can generate interface code in minutes —a task that used to take half a day, Li said. Currently, more than a dozen internal projects at Ant, including PC websites, mobile apps, and mini-programs, use Graph-to-Code, producing hundreds of thousands of lines of code. CodeFuse’s internal adoption at Ant has reached 89%, with younger programmers leading its usage, Li said.
In October 2023, Alibaba Cloud launched its AI-assisted coding tool Tongyi Lingma
In October 2023, Alibaba’s own cloud unit launched its AI-assisted coding tool Tongyi Lingma, supporting 16 programing languages and progressively adding features such as real-time code continuation, optimization, code comment generation and unit test creation.
In April, Alibaba Cloud said it expected 20% of its code would be written by Tongyi Lingma, allowing programmers to focus on system design and core business development. In September, Alibaba Cloud said that Tongyi Lingma had evolved from handling basic coding tasks to completing a full range of development tasks — such as requirements understanding, task breakdown, coding, bug fixing and testing — through simple natural language dialogue.
Internet companies are not only promoting AI productivity tools internally to increase efficiency but are aiming to scale their AI capabilities to gain market leadership.
Baidu’s Zang said that since adopting AI assistants, programmers have increased their monthly output by more than 10% compared with two years ago. “For companies with thousands of programmers, this improvement draws the attention of chief executives and finance officers,” he said, adding that some 10,000 companies are now testing Comate.
Alibaba Cloud said its Tongyi Lingma has already been adopted by companies such as China United Property Insurance Co. Ltd, Hello Group Inc. and Changan Automobile Co. Ltd., collectively generating more than a billion lines of code. With skills across roles such as architect, developer and test engineer, Tongyi Lingma has been widely used in sectors such as finance, manufacturing, internet, transportation, automotive and energy.
New normal for programmers
A study by Ant Group and the Bank for International Settlements (BIS) found that programmers using the AI assistant CodeFuse boosted their code output by an average of 55%, with about a third of this increase coming from AI-generated code.
The study showed that productivity gains were concentrated among junior programmers. Ant’s Li told Caixin that a monthly output threshold of 4,000 lines of code marks a difference: junior engineers below this level benefited most, as CodeFuse helps them complete code while they plan. In contrast, experienced programmers, who generally exceed this threshold, often find manual coding faster due to their expertise and familiarity.
Other BIS research highlights AI’s productivity impact across jobs. Data from the New York Fed shows that generative AI improves productivity by 18% for customer service roles and 60% for business document writers, while programmers using GitHub Copilot complete more than twice as many projects weekly.
The impact of AI on programmer demand is raising concerns. Zang said that companies are currently focused on efficiency gains rather than cost-cutting in R&D. Full end-to-end delivery by AI remains out of reach, making experienced programmers essential for process design, with coding as just part of their role.
An IT department head at a bank agreed. “AI is enhancing the programmers’ experience rather than replacing them. With AI handling some code, programmers can take breaks or tackle more requests,” he said.
Many predict that AI could significantly disrupt the labor market. “Large models may eliminate entry-level jobs for young people,” said He Fan, a professor at Shanghai Jiao Tong University. He compared AI’s current progress to the steam engine’s readiness in the mid-18th century — decades before trains emerged. When major AI applications finally arrive, he warned, the impact will be immense.
Some roles and groups are expected to be more vulnerable. Research by Accenture, the professional services company, shows that more than 40% of work hours in China’s retail, life sciences, high-tech and energy sectors could be optimized or automated by generative AI. Employees with limited digital skills, less experience and lower education levels are at higher risk, raising concerns about an expanding digital divide.
The future of programming
Li compares AI coding assistants to self-driving technology, suggesting they are currently at an “L2.5” level, capable of handling specific tasks like front-end design, testing and debugging within limited scenarios. Reaching “L3” capabilities, where AI can complete end-to-end generation in certain contexts, could take three or more years, while full “L4” automation, where AI completely replaces human programmers, remains a distant prospect, Li said.
The potential of advanced AI, however, raises questions about the future dynamic between programmers and technology, creating both career uncertainty and anxiety.
A Google algorithm engineer said that Silicon Valley’s top companies are racing to release even more powerful models, including OpenAI’s GPT-5 and GPT-6 and Anthropic’s Claude 4 and Claude 5, all expected in 2025. “It’s only a matter of time before programmers are replaced by AI,” he said.
One tech employee envisioned that a company’s most valuable asset might eventually shift from talent to authentic data, with raw information AI cannot generate becoming the new “oil.”
To address these concerns, tech companies are positioning AI coding assistants as tools to support rather than replace human roles. Baidu has described Comate as a product aimed at empowering programmers.
“Last year, we debated whether AI would replace humans; now, we see the greater challenge is deploying AI effectively and practically,” said Zhang.
Feng Yiming contributed to the story.
caixinglobal.com is the English-language online news portal of Chinese financial and business news media group Caixin. Global Neighbours is authorized to reprint this article.
Image: Akira Kaelyn – stock.adobe.com