Association Responds to 'AI Replacing Programmers': Can an Official Stance Quell Market Panic?

Opening: An Official Association Steps In for the First Time, Yet the Comment Section Argues Even More Fiercely

On September 11, 2026, the Ministry of Industry and Information Technology held a press conference for the “Artificial Intelligence + Software” Special Action Implementation Plan. The plan itself was the main topic, but a question from a reporter on site stole almost all the attention: intelligent programming is becoming increasingly popular, software practitioners are under significant pressure, and the social discussion that “artificial intelligence will replace programmers” is spreading fiercely—what is the official view? Answering this question was Lu Weifeng, Vice President and Secretary-General of the China Software Industry Association. The Beijing News report used the very first sentence of his response: “This issue has high social attention, and ‘we also attach great importance to it’.”

These words spread much faster than the plan itself. Starting at noon that day, groups where programmers congregate were forwarding this news, and the term “Association responds to AI replacing programmers” hung on the trending list of career communities all day. For many people who write code, this is the first time an official industry association has formally stepped forward to take a stance on “whether AI will make me unemployed”—in previous discussions, those speaking up were either AI company CEOs or course sellers; either trying to scare people or comfort them, but no one needed to be accountable for what they said.

But if you read this response carefully, you will find that it never says “it will not replace,” not even a single sentence. It talks about the adjustment of capability structures, the reconstruction and addition of jobs, and three things the association will do next. This is worth pondering: is the official here to give a reassurance pill, or to admit that changes are happening using more decent wording? In this issue, we will break this down.

Shiwen: An official association has stepped in to respond to “AI will replace programmers” for the first time. What was your first reaction—a reassurance pill, or an official announcement?

Yongliang: Neither. My first reaction was: the official rhetoric has finally caught up with the frontline experience. And the order of appearance is intriguing—the plan came out first, then the panic was addressed at the press conference, which shows that the pressure in the industry has become too big to hide.

Shiwen: Then let’s talk it through thoroughly: what exactly did this response say, why does the official rhetoric clash with the feeling in the recruitment market, where is the statement “AI replacing programmers” wrong, how have your own team’s hiring standards changed, and what can ordinary programmers do now?

Q1: What exactly did this response say, and how much weight does it carry?

Yongliang: The value of this response lies not in the conclusion, but in the fact that it wrote the word “change” into the official text openly and properly, without glossing over it.

I look at it in three layers. The first layer is the stance: “attach great importance.” First, I need to pour a basin of cold water here—industry associations have no regulatory power and cannot order any company not to lay off staff. Their “importance” is about leveling the topic, not a policy promise. Hearing the words “attach great importance” and thinking your rice bowl is secure is a misreading. The second layer is the diagnosis, which is the most solid part of the whole text: three points on talent capability structure—the weight of pure coding execution is declining, while the weight of requirement understanding, architecture design, system integration, and quality and safety control is rising; human-machine collaboration is becoming a new basic capability, and accurately expressing development intent, evaluating and verifying generated results, and orchestrating multi-agent collaboration have become essential skills; composite talent is obviously scarce, and new roles like frontline deployment engineers have even emerged in the industry, requiring both technical understanding and soaking in customer sites to understand the business. Two points on job structure—existing jobs face reconstruction, while incremental jobs continue to emerge. The most critical sentence in the middle is: “The incremental jobs created by new technologies are not a simple equivalent replacement for stock jobs; the pressure on repetitive basic work posts and the shortage of composite talents may coexist for a period.” Translated into plain language: the official admits that some people will be squeezed out by this wave of changes, and it’s not as easy as just changing seats. The third layer is action: three things—job reform and skills training, talent supply and supporting practitioners in transitioning to high-value-added positions, and cultivating new business formats like intelligent agent software.

How do we judge the weight? I have been in software for 17 years and have seen many statements from industry associations; most are correct but harmless nonsense. The difference this time is that it did not deny the change, but outlined the specific shape of the change. It is equivalent to telling all companies: in the future, when carrying out intelligent transformation, the placement of people must be put on the table; it is no longer a little secret for internal digestion. Panic has been upgraded from a self-media topic to an industry governance topic, and this is the real weight of this response.

Q2: The officials say “attach great importance,” but in the recruitment market, are junior jobs really still there?

Yongliang: Every sentence the association says is correct, but when I open the recruitment backend for 2026, there is a real crack between the experience and the rhetoric—this crack is the place we should watch most in this issue.

First, the official rhetoric: capability weights are adjusting, jobs are being reconstructed and added, the Ministry of Human Resources and Social Security has included occupations like Generative AI System Operator in the national occupational classification dictionary, and added new types of work like Industrial Software Designer and Generative AI System Tester. Now, my frontline experience. In the first half of 2026, my team released three recruitment slots: one senior engineer, two juniors. The senior position received over four hundred resumes; it was fierce competition, but I could find gold. The situation for the junior position was heartbreaking: about 70% of the resumes received carried a strong trace of training bootcamps—the projects were uniformly shopping malls and backend management systems; ask “how did you handle this concurrency scenario,” and few could answer. In 2023, opening the same junior position, I could pick out a dozen usable ones from a hundred resumes. In three years, the ratio that passed the initial screening dropped by an order of magnitude.

So have junior jobs disappeared? No, they have gotten more expensive. Before, recruiting a junior meant recruiting “cheap hands,” as long as they could do repetitive work; now recruiting a junior means recruiting “someone who tunes AI”—when a requirement comes, they must first know how to break it down, feed it to the tool, and verify it after feeding. This requirement directly blocks a large number of candidates who just want to “write code for a salary.” This is the reality version of “coexist for a period” in the official document: between the people being squeezed out and the threshold of new jobs, there is a distance of one to two years of retraining; this distance is called “transition period” in the document, and “gap period” for the individual. The most uncomfortable are those with three to five years of work experience whose skills are stuck at “proficiently writing business code”—upward, the pits for architecture and requirements are occupied by more senior people; downward, newcomers are cheaper and willing to learn. The official rhetoric looks at the total plate of the whole industry, while individual experience looks at the crack in front of them. Both perspectives are correct, but the individual cannot wait for the curve of the total plate to slowly improve.

Q3: Where exactly is the statement “AI replacing programmers” wrong?

Yongliang: It lies in swapping the unit—AI replaces tasks, not professions, and panic equates these two units directly, then multiplies it by a family’s income.

Let’s break it down. The profession of “programmer” was never about just one line of code: understanding requirements, weighing solutions, coding, joint debugging, going online, carrying the pot in the middle of the night—it’s a whole chain of life. What AI really eats is the segment of “translating a finalized plan into code”—which happens to be the main output of junior positions in the past, and also happens to be the link most easily standardized. So the precise version of “AI replacing programmers” should be: “AI ate the segment of the programmer profession that is most suitable for standardization and most suitable for newcomers to practice.” Is this difference important? Too important. If the profession were gone, then everyone would have no choice, and lying flat would be reasonable; if a segment of the profession is gone, then it is a capability reorganization, and there is a choice.

There is a second error in this statement: it assumes a position is a static stock, one goes out only when one comes in. In fact, the metabolism of software positions has never stopped; it’s just that before, the metabolism was about the tech stack, and this time, the metabolism is about the link. The third error is more subtle: it ignores cost transfer. Before, a company hired three juniors to write CRUD (create, read, update, delete), now it hires one senior to lead AI to do the work of five people—on the macro ledger, productivity has risen, but on the micro level, the opportunities for those two juniors are gone, and the loss of these two people will not appear in any “industry overall improving” statistics. So this statement cannot be simply said to be wrong; it is half right: what is being replaced is not “programmers,” but the persona of “programmers who only know how to write code.” The official response actually said this very cleanly—the weight of coding execution is declining, and the weight of requirement understanding, architecture design, and quality and safety control is rising. You savor it; not a single word in these sixteen characters says “it’s okay.”

Q4: As a technical director with 17 years of experience, how have your hiring standards changed in 2026?

Yongliang: They have changed, and they changed thoroughly in less than two years—my screening logic has changed from “looking at what he knows” to “looking at how he commands AI and how he verifies AI.”

Let’s talk about three specific changes. The first, the written test questions have changed. Before, I tested algorithms and hand-coding; now, I give a real small requirement on site, equipped with AI tools, ninety minutes, and I only look at three things: what does he ask first when he gets the requirement—this tests judgment; how does he verify what the AI generates—this tests acceptance ability; how does he adjust when the AI answers irrelevant questions—this tests the ability to express intent. The two engineers recruited with this question last year are now the main force in the group, proving that this question is effective. The second, resume screening for junior positions added a “AI trace check.” Those whose project descriptions are full of beautiful AI-generated jargon and who freeze when asked “why did you write this code this way” are passed directly. It’s not that I’m against using AI; I use it every day, but I need to distinguish between “he knows” and “AI knows”—these two things look exactly the same on a resume, but they are revealed as soon as you chat. The third, we are willing to pay a premium for “human-machine collaboration capability.” For the same years of experience, someone who can break down a set of tasks for several AI agents to run in parallel and then verify them one by one gets a salary half a grade higher. To put it bluntly, the labor market has started to price this ability explicitly.

There is also a set of numbers I will say for everyone to weigh: my team peaked at 70 people, now it is 52, but the output has not dropped. It’s not because 18 were fired—these two years were natural attrition without backfilling according to the original establishment, and the empty workload was eaten by AI and process optimization. The hardest change to accept is this: I can no longer give newcomers the space to “practice slowly.” Before, the master led the apprentice and fed them real business; now, the part of the work suitable for feeding the apprentice is exactly what AI does, and newcomers have nowhere to practice. This is an industry-level problem. We make up for it internally with special training of “AI generation, human bug-finding,” but I know this is not a long-term answer.

Q5: Without selling anxiety, what should ordinary programmers do now?

Yongliang: Four things, do them in order, and each can get feedback within three months.

The first thing is to remove “coding execution” from your main label and replace it with a business label plus an AI collaboration label. Writing “three years of Java” on a resume has no recognition; writing “three years of experience in medical informatization, able to use AI tools to complete the full process from requirement to launch” is immediately different. The composite talent the association talks about is exactly the kind of person the market is willing to pay a premium for. The second thing is to take the sentence “accurately express development intent, evaluate and verify generated results” from the official response as a new job description to practice—translated into daily actions, it is: before taking a job, write clearly what is wanted, what is not wanted, and what counts as success; before handing over the job, verify these three items one by one. This is the practice method with the lowest cost and fastest effect; I can see the gap in my team’s newcomers in three months. The third thing is to actively use AI in your most familiar business deep waters. General coding is known to everyone, and its gold content approaches zero, but the industry know-how you have soaked in for five years is the moat—the new role of frontline deployment engineer in the association’s response earns exactly this money. The fourth thing is to keep an eye on the Ministry of Human Resources and Social Security’s new job catalog and the association’s follow-up training actions. You don’t have to go get a certificate, but that is the most direct wind vane for job migration. Knowing half a year in advance which way the wind blows versus knowing afterwards is two different lives.

Finally, a truthful word: don’t wait for the official reassurance pill. What the association’s response can do is turn “transition” into a public issue and increase the social pressure on companies to place employees; but personal transition has always been self-funded, self-stayed up late, and self-carried gaps. Panic will not disappear because of an official statement; it will only disappear at the moment you have new things in your hands.

Epilogue

Shiwen: Finally, summarize this issue in one sentence?

Yongliang: The association did not say “AI will not replace programmers,” it said “the profession of programmer is changing its kernel”—the exit of the old kernel is a fact, and the opening of the new kernel is also a fact. Which side you stand on does not depend on any statement, but on what you practice in the next six months.

Shiwen: This sentence is for everyone. See you next time.

---

[Technical Depth] What the Association Can and Cannot Do: The Boundaries of the Three Grips of Training, Standards, and Statistics

This issue discusses “whether an official tone can stop panic,” and for readers who want to go a step further, I will break down a layer: does the subject of the industry association have cards in its hand? The answer is yes, but the scope of each card is limited. Only by seeing the boundaries clearly can we know how to use this response correctly.

The First Grip: Training and Certification. What the association can do, and is best at, is translating “human-machine collaboration capability” into courses, certifications, and practical training projects, giving people in transition a path with signposts. The “simultaneous implementation of job reform and skills training” mentioned in the response falls to the operational level mainly relying on this set of things. But its boundary is also clear: certification can only prove that a person has learned, not that he can use it. What the market ultimately recognizes is the quality of the work; training solves “knowing,” not “doing.” Expecting a certificate to offset unemployment risk is the wrong direction.

The Second Grip: Industry Standards. This is the association’s real hard tool. It can write the quality requirements of intelligent programming tools and the personnel placement requirements in the process of enterprise intelligent transformation into industry standards and group standards, making “placing people” change from a slogan to an item in the acceptance of large customer procurement and member units. This move has teeth for companies of a certain scale. But the boundary is equally obvious: standards can constrain member units and formal procurement, but cannot reach a large number of small companies and the outsourcing market—and junior positions are highly concentrated in these places where standards do not shine.

The Third Grip: Employment Statistics and Public Appeals. The association has the conditions to continuously publish job structure data, exposing the scale, industry, and regional distribution of “periodic coexistence” with numbers, using facts to suppress rumor-style panic; it can also continue to shout through the media and press conferences to change the temperature of the public opinion field. The boundary of this set of actions is time: statistical data naturally lag behind the market by one to two quarters, and by the time the report comes out, the individual’s gap period has already actually occurred. Statistics can stop rumors, not feelings.

So returning to the question at the beginning: can an official tone stop the market’s panic? My judgment is—it can stop the part of rumors, but not the part of feeling. It can change the level of the topic and the cost of excuses for companies, but it cannot change the one-to-two-year retraining threshold in front of everyone. Seeing these three boundaries clearly, the correct way to use this response becomes clear: treat it as a roadmap, not a painkiller.

---

Information Sources for This Issue

• Beijing News 2026-09-11 “Will AI replace programmers? China Software Industry Association responds”
• IT Home 2026-09-11 “China Software Industry Association’s Lu Weifeng responds to ‘AI replacing programmers’: attach great importance” (Key points from the press conference transcript)
• The response by Lu Weifeng, Vice President and Secretary-General of the China Software Industry Association, at the Ministry of Industry and Information Technology’s “Artificial Intelligence + Software” Special Action Implementation Plan press conference

=== END ===