This time, it's really not "The Boy Who Cried Wolf": Under the dominance of AI, the job market for programmers has been completely reshuffled.

Exactly one year ago, I had a meal with the person in charge of a mid - sized Internet company. The person told me that thanks to AI programming tools, they had laid off one - third of their software development team in the past two years, and might lay off another one - third in the next few years. This constituted the main achievement of the company's "cost reduction and efficiency improvement" strategy, making a significant contribution to its financial performance. He didn't tell me which AI programming tool the company was using, but I guessed it was probably Cursor. At that time, and even until today, Cursor has been the most popular AI programming tool in the world.

The person said that those who were laid off first were often "mid - level" programmers with three to five years of work experience. As their seniority increased, so did their salary requirements, but it was almost impossible for them to make substantial improvements in their abilities. They had only accumulated experience and proficiency, traits that could easily be replaced by AI. The most outstanding programmers couldn't be replaced, and their salaries wouldn't be cut. Newly - recruited junior employees, with low salaries and great potential for development, didn't need to be replaced immediately. In short, the company's strategy was to lay off (or not renew the contracts of) those "old timers" who only had seniority and replace them with young people. As for the lack of experience among young people, AI could handle it quite well.

This has actually been the consistent strategy of the entire Internet industry and even the global technology industry in the past few years: using AI to replace human labor in standardized and repetitive programming work, especially "mid - level" labor with a relatively low cost - performance ratio. However, there has been controversy in the public about the sustainability of this strategy. I remember that in July 2025, someone bought a trending topic on Weibo: "The idea that AI will replace programmers is the fantasy of laymen." I'm also a layman when it comes to programming, but the moment I saw this trending topic, I sensed that someone was clearly anxious. It turned out that the topic was bought by a programming training institution. No wonder!

Strictly speaking, until a week ago, the scope of "AI replacing programmers" was limited and controllable. A large amount of work still needed to be done manually or at least continuously supervised by humans. However, two products launched in the past week have completely changed the situation: Claude Code (powered by Claude Opus 4.6), and GPT - 5.3 - Codex.

Here are the comments from some of my friends who work as programmers or have programming as a hobby:

"It won't surprise anyone that programmers will face large - scale unemployment. Both tools are very powerful! The future employment trend is clear. Junior programmers who just fill positions will no longer be valuable. And no one knows if someone will launch an even more powerful tool tomorrow, accelerating the process."

"Programmers have to face unemployment, including me... The two tools each have their own advantages, but one thing is certain: most programmers are really no longer valuable. Only God knows what the future holds."

"It's actually a good thing for entrepreneurs. As long as they have a general understanding of programming principles, they can save a lot of costs in building a development team. It's a pity that given the current economic situation, not many people dare to start a business."

I can't help but recall that not long after ChatGPT was launched three years ago, there was a popular question on Zhihu: "Why does generative AI only impact liberal arts and not science and engineering?" It was the stupidest question I've ever seen on Zhihu. Facts have proved that generative AI impacts every industry equally, whether it's liberal arts, business, science, engineering, or medicine. Everyone's comfort zones have been shaken. Those who realize the crisis earlier have a chance to prepare earlier. On the contrary, those who indulge in the illusion that "AI only impacts others, not me" are probably like the passengers on the Titanic right after it hit the iceberg, not knowing they're doomed until the very end.

Now, let's talk about some technical details: What makes GPT - 5.3 - Codex and Claude Code so epoch - making? In a nutshell, they are true "agents", capable of developing application programs throughout the entire process and even automatically, although they have different focuses.

Cursor is just a pure tool without a base large - language model. It achieves programming by calling other large - language models. So far, the most commonly used large - language models by Cursor users are the Claude series. Cursor has always been positioned as an "editor with AI as its featured function". You can type a few lines of code on it, and it can efficiently predict the next few lines and modify existing code. The most important reason for users to use Cursor is that it can clearly display the code created and modified by AI and efficiently maintain the workflow and code library. After all, it's a tool to "help programmers improve efficiency". Although "improving efficiency" itself means that many programmers will lose their jobs, its goal isn't to replace programmers.

Claude Code and Codex are intelligent agent tools aimed at replacing programmers. Both have strong programming capabilities, with slightly different focuses:

Claude Code has strong in - depth reasoning ability and is good at handling complex program architecture problems. It supports an ultra - long text context window (up to 1 million tokens). However, its automation level is slightly lower, and its generation efficiency is significantly slower than that of Codex.

Codex focuses on agentic execution, with a high degree of automation and fast generation speed. However, its in - depth reasoning ability is slightly weaker, and it only provides a context window of 400,000 tokens.

Therefore, some people call Claude Code a "Thinker" and Codex a "Doer". In fact, the difference between the two isn't that big. Both can complete complex application development tasks in an almost automated way, only with different design ideas and innovation directions. For example, one of Codex's features is the "Steer Mode", which allows users to interrupt code generation and make modifications midway through a task. While this is beneficial for combining human brains with AI, it also means that the completeness of the code generated by Codex in one go is slightly weaker than that of Claude Code. As time goes by, future versions of these two products will surely become more comprehensive and meet the needs of different users.

Don't forget that there's a new player eyeing the market and ready to enter at any time - the rumored new version of DeepSeek. Judging from the previously published papers, the key ability that DeepSeek will focus on improving in the next stage is also programming. If it also adopts the automated approach of an agent, we're likely to see three epoch - making "programming agents" within a month. This will be a moment of celebration for the AI industry, a time for the capital market to rejoice, but a time of complete reshuffle for millions of programmers around the world...

Will the software outsourcing (ITO) industry still exist in a few years? You know, the "Indian homework - doing - for - pay" industry represented by Chegg has been destroyed by AI in just two years. I remember that more than a decade ago, when I first started researching the software industry, software outsourcing was a trendy field. Everyone hoped that "Beijing would overtake Bangalore" and become the world's programming base. Now it seems that Bangalore itself is in trouble, and probably no city will be interested in overtaking it.

For all companies whose main business isn't software development, their development teams will be further downsized. In the past, these teams were often outsourced to third - parties. Now, they may still be outsourced, but the final executors have changed from humans to AI. With the "simplification" of AI programming tools, perhaps the outsourcing process can be omitted, and a small but elite in - house AI development department will suffice.

What about large Internet companies? They may be the quickest to adapt to the new era and the most active in laying off employees because their middle - and senior - level managers are the most tech - savvy, and most of them are former programmers. Who knows better than programmers themselves how to lay off programmers? As the saying goes, "What makes you thrive will also make you decline." But there should be a second half to this saying - it was the past self that thrived, and it's the future others that will decline.

Anyway, in this era, no one has time to be sentimental. Just a few days before the launch of Claude Code and Codex, Google launched Genie 3, a "world - class large - language model", to its top - tier paying users, which triggered a panic sell - off of game companies and game engine developers in the capital market. Ironically, Genie 3 was actually developed as early as August last year but hadn't been opened to users. What Wall Street is experiencing now is the panic that should have occurred six months ago.

A few days later, ByteDance launched its large - language model, Seedance 2.0. Fairly speaking, this is the best video - related large - language model launched in China so far and one of the best in the world. Many of my friends spent the whole weekend fine - tuning Seedance and evaluating its impact on the film, television, and video production industries. As for which industries' practitioners will be affected, that's a topic for another article.

Are the barbarians at the city gates? No, the barbarians have already invaded Rome and are walking the streets fully armed, asking the Romans, "Do you want to coexist with us or be eliminated by us?" Optimistic people are preparing for centuries of coexistence, pessimistic people are preparing for a last - ditch battle, and stubborn people are still repeating nonsense like "The idea that AI will replace programmers is the fantasy of laymen" and "AI will never revolutionize the film and television industry".

No matter what happens, we can only stride forward in the direction of the storm. Because this is the only way to survive and grow again.

This article is from the WeChat official account "Internet Phantom Thieves Group" (ID: TMTphantom). Author: Pei Pei, the leader of the Phantom Thieves Group. It is published by 36Kr with authorization.