With the rapid development of AI technology, AI programming tools are gradually changing the way programmers work. From code completion to semi-automatic programming to fully automated programming, the capabilities of AI programming tools are constantly improving, raising concerns among programmers about their career future. This article delves into the development status of AI programming tools, their impact on the programmer community, and the future transformation of programmers’ roles.
The wind of AI replacing humans is accelerating to the programmer community.
In the past two years, the emergence of AI tools such as ChatGPT and Midjourney has made professional groups such as copywriters and illustrators tremble, and at this moment, programmers are also in the anxiety of being replaced by AI.
Especially recently, American AI unicorn company Anthropic released a newly upgraded large model Claude4 series, which once again made programmers around the world feel pressured. The series includes Claude Opus 4 and Claude Sonnet 4, the biggest feature is the programming time and understanding ability, especially the Claude Opus 4 can continue to write code for 7 hours, known as “the world’s first large model that can generate high-quality code without manual modification”.
Judging from the data, the popularity of AI programming tools continues to rise. According to the latest statistics from data company Xsignal Singularity, AI programming (AI R&D tools) ranks third among more than 30 AI application scenarios, surpassing popular applications such as AI search engines and AI image generation. From June 2024 to April 2025, the AI tool’s social media discussions also increased by 45%.
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
At the same time, the actions of large technology companies have also released subtle signals. Microsoft recently announced that it will lay off 6,000 employees worldwide, involving core technical positions such as engineering and R&D, which has become a landmark event under the impact of AI.
Whether it’s the continuous iteration of AI programming tool capabilities, the rapid growth of user usage data, or the intensive adjustment of large technology companies, they are constantly exacerbating programmers’ concerns.
How far have AI programming tools come from? Can it really replace programmers entirely? An unresolved torture of the times is spreading among programmers.
01 Who is the strongest AI programmer in the world?
The AI programming track is ushering in a “100-model war”.
In addition to the Claude4 mentioned above, the active AI programming tools on the market are now “blooming”, and both large companies and entrepreneurial teams are unwilling to give up this track.
Based on the end-of-period sound volume value (i.e., the number of times it is mentioned on the Internet, which can reflect the popularity of social media discussions) and the perception of practitioners provided by Xsignal Singularity Factor, “Fixed Focus One” sorts out the top ten AI programming tools with high popularity at home and abroad.
It should be pointed out that although the final volume of the Kimi-AI programming assistant is very high, it is not a standalone product, but a module integrated into the Kimi application. The statistics show Kimi’s overall final voice value, and the actual degree of discussion is much lower than this number.
The relatively popular AI programming tools in China are basically dominated by large manufacturers, such as Alibaba’s Tongyi Lingcha, Baidu’s Wenxin Kuaicha and Byte’s Trae.
Their commonality is that they are more prominent in one or several aspects such as the threshold for use (converting natural speech into code) and intelligence (automatic completion and detection of code).
Abroad, giant companies and entrepreneurial teams have good products.
On the one hand, Microsoft’s GitHub Copilot not only supports multiple programming languages, but also integrates seamlessly with GitHub’s codebase. On the other hand, Cursor, created by the entrepreneurial team Anysphere, has risen rapidly, not only “complete” code, but also generated, repaired, and understood code, and has become a star product among AI programming tools.
From the perspective of user activity (MAU), Cursor has ranked among the top in the world in March, and domestic Trae, Tongyi Lingma, and Wenxin Kuaicode have also ranked among the first echelons. GitHub Copilot does not disclose MAU, but it is also at the forefront of the industry in terms of practitioner perception.
Qin Xiang, an AI software engineer, said that AI programming tools provide convenience for developers in lowering thresholds, improving productivity, promoting innovation, and optimizing complex systems, whether at home or abroad.
The development path of AI programming tools is roughly divided into three stages: from code completion, to semi-automatic programming, and then to fully automatic programming. At present, most AI programming tools on the market are semi-automatic programming tools represented by Cursor and MarsCode, and developers will check and adjust the generated code, which has the advantage of significantly improving efficiency while retaining human dominance, while fully automatic programming mainly serves novice users.
So, how to judge the strength of an AI programming tool? The comprehensive experience of practitioners can be measured from two dimensions: technology and function.
On the one hand, the technology of AI programming tools relies on the large model capabilities behind them. Senior programmer Lu Tong said that the underlying technical principle of AI programming is large language model + code-specific training optimization, and the large models that are more suitable for AI programming in China are DeepSeek and Qwen series, and foreign ones are Claude, Gemini, GPT4, and the current Claude series models are considered the most suitable for AI programming because of their code understanding and long text processing capabilities.
On the other hand, it also depends on the ability to handle complex development processes, such as the ability to understand multiple code files, fix bugs, generate front-end interfaces, generate code based on UI image recognition, and independently call tools and operating system commands. The more and smoother the processes that can be automatically processed, the stronger the level of the AI programming tool. Cursor is an example, which can autonomously complete the entire process from requirements to feature development, and is used as an assistant by many developers.
The continuous advancement of AI programming tools has made programmers excited about the dividends brought by efficiency improvement and the risk of being replaced. The next question is: what type of programmer will be replaced first?
02 AI programming, replacing junior programmers is not a legend
If we compare the performance of AI programming tools to programmers, most of them have reached or even exceeded the level of junior programmers, and even some products have the ability of intermediate programmers.
Practitioners introduced to “Fixed Focus One” that from the perspective of job division, programmers can be roughly divided into different directions such as front-end, back-end, full-stack, embedding, etc., of which each category is subdivided into beginner, intermediate and advanced, the main difference is the depth of participation in the development of products and the number of technical principles mastered.
Junior programmers are usually only responsible for developing simple functions, such as adding, deleting, modifying and checking the content of the system, and do not require too many other abilities. Intermediate programmers need to master most of the code technical principles and be responsible for relatively simple function development, such as interface and database design; Senior programmers are responsible for the technical selection, framework construction, core algorithm design, and functional development of core modules of the entire system.
At present, many AI programming tools are not limited to generating code, but can also achieve the whole process of early idea framework organization, intermediate code generation, and post-optimization interface, which has surpassed that of junior programmers in terms of competency. Lu Tong said that many product managers who do not understand code have become independent developers through AI programming tools, and some senior programmers have replaced manual labor with AI programming and become their left and right hands.
He gave the example of developing a psychological testing app.
Although the application products are small, but the development difficulty is not low, the psychological testing application should not only ensure the diversity of psychological test types, but also consider user privacy protection and data security, which involves front-end and back-end development, database management, API interface and other functions, which require beginner, intermediate and senior programmers to cooperate in different links, and AI programming tools can almost intervene, the process includes:
Step 1: Let AI recommend more popular psychological testing applications, such as MBTI psychological tests, personality color tests, and career matching tests;
Step 2: Generate the specific functions required for the psychological test application, such as login, registration, psychological test question display, answering, and sharing;
Step 3: According to the determined functions, draw their interface sketches with the help of AI;
Step 4: Let the AI generate interface sketches and the required code behind the functions It should be pointed out that many AI programming tools also support selecting specific technical frameworks;
Step 5: Run the resulting code and use AI to adjust and optimize the functionality and interface until it meets the requirements.
It is not difficult to find that from product conception to function implementation, as long as users can logically describe their needs and use natural language throughout the process, AI programming tools can achieve it, greatly lowering the development threshold.
More than one programmer said that AI programming is becoming more and more capable and they use such tools every day. Lu Tong most often uses Cursor and Tongyi Lingcode, and he basically programs through prompt words to interact with AI, and he does not type code line by line.
Qin Xiang added, “Cursor has obvious advantages in cross-file development efficiency; The Chinese optimization and privatization deployment capabilities of Tongyi Lingcode are outstanding; Claude 4 can handle complex tasks and is suitable for full-stack development. ”
Lu Tong introduced that using AI programming tools to develop applications can save nearly half of labor costs and time. Some programmers said that if they are proficient in using AI programming tools in their daily work, their work efficiency can be increased by 30%-40%.
The efficiency improvement stems from the high technical compatibility of AI programming languages with large models. Lu Tong introduced that there are few code keywords, and the standardization of programming languages and AI are very compatible. Taking Cursor as an example, it can not only generate complex code, but also solve code errors independently during the debugging process, and also support the modification of global or partial code.
However, at a time when AI programming capabilities are rapidly evolving, does it mean that all programmers will be replaced? What kind of choice will the company behind the programmer make?
03 True programming ability is being redefined
A harsh reality is that the rapid evolution of AI programming tools has begun to affect the employment stability of programmers.
In May this year, a piece of news shook the industry: on May 13, Microsoft announced a major layoff involving 6,000 employees around the world, with programmers bearing the brunt of many positions. According to media reports, about 41% of Microsoft’s 2,000 laid off people in Washington State are related to software engineering positions, including senior engineers like Ron Buckton, the core developer of the TypeScript compiler.
While Microsoft does not directly attribute the layoffs to AI replacing manpower, its investment in AI programming is a signal. CEO Nadella has revealed that more than 30% of the company’s code is generated by AI, and CTO Scott has also predicted that by 2030, this proportion will exceed 95%. AI code has not only penetrated Microsoft, but its competitor Google has also revealed that more than 25% of new code is generated by AI.
Although there has been no large-scale news of programmers being replaced in China, practitioners have already felt the crisis. Lu Tong said that AI programming tools have developed faster than he imagined.
He recalled, “In 2023, I think AI is only an auxiliary programming tool, and AI programming tools such as Tongyi Lingcode and Cursor can only generate part of the code. By the end of 2024, after the launch of Cursor’s agent mode and Tongyi Lingma’s ‘AI programmer’ function, it has been able to realize functions such as independent multi-code file generation, automatic reading of project files, and automatic start and execution of code. ”
He believes that AI programming tools have caught up with full-stack development senior programmers in terms of comprehensive functions and work efficiency.
Even so, many programmers still believe that AI is currently more of an efficient assistant than a complete replacement for programmers. If you want to completely replace this profession, there are at least three levels to pass in AI programming.
First of all, the understanding is insufficient, and it is difficult to accurately “understand” complex needs.
AI writes code quickly, but its understanding needs to be strengthened. Lu Tong mentioned that when he uses AI programming tools to modify the front-end code, he needs to adjust the prompt words many times to get satisfactory results, which is also a common problem of all AI tools – high requirements for prompts. If the input logic is slightly ambiguous, the result will be off.
Secondly, there is no subsion for complete product development thinking and team collaboration.
In software research and development, writing code is only one of the links, and it also involves product requirements research, innovative design of product tools, etc., which is also a necessary ability for senior programmers, which is not possible with current AI programming tools.
Qin Xiang believes that although AI programming tools have moved from “basic completion” functions to “semi-automated collaboration” advanced functions, and head tools such as Claude 4 and Cursor are still upgrading towards “fully automatic programming”, the core role of humans in architecture design and business understanding is still irreplaceable.
Lu Tong said that if there is no very complex business process, for example, the product developed is a C-end tool or SaaS application, with a clear process and a standard structure, AI can do most of the work. On the one hand, it is difficult for it to understand a company like a human, and on the other hand, there will be over-analysis, random changes, and even frequent restructuring of code, which will affect the stability of the project.
Finally, AI programming has an extremely low fault tolerance rate.
If there are problems such as grammar and picture errors in AI-generated text and pictures, users may be reluctantly acceptable, but once the code goes wrong, it will fail at best, and lead to safety accidents at worst. Although AI programming tools can already automate the monitoring of the generation process, there are also problems such as “the generated code seems complete, but it does not consider compatibility with the operating system or browser”. A programmer said that sometimes, AI code “looks right” does not mean “it runs right”. In addition, if the AI-generated code has security vulnerabilities, it is difficult to clearly define the attribution of responsibility. This is still an unresolved compliance issue in the implementation of technology.
Therefore, for AI to truly replace programmers, it faces not only technical challenges, but also multiple thresholds of understanding, creativity and responsibility.
AI programming won’t put programmers out of work overnight, but it’s reshaping the core values of the profession. Future programmers may no longer mechanically “knock code”, but understand both AI and business, and assume higher-dimensional capabilities.
