"I would rather be unemployed than write a single line of code with AI," a 20-year industry veteran announced that he will stop using AI after relying on AI tools for 18 months, as he wants to regain the joy of writing all code manually from scratch.

If someone tells you that AI can make programmers write code several times faster, you probably won't find it surprising.

But if a veteran coder with over 20 years of coding experience tells you: "I've been using AI for more than a year, and I do write code faster now, but I'm growing to hate programming more and more, so I've decided to stop using AI completely." then this is something worth talking about.

Recently, a software developer named Brett shared his personal experience.

He has been coding for more than 20 years, and has long used AI programming tools such as Cursor and Claude Code in the past year and a half. According to most people's imagination of AI programming, he should have been one of the biggest beneficiaries: completing more tasks, producing code faster, and even once feeling that he was working in "the way of the future".

However, in the end, he made a seemingly very counter-trend decision: stop using AI programming tools, and even stop using AI in his daily life. The reason is not that AI is not powerful enough. On the contrary, it is precisely because it is so capable — when AI does more and more of the programming work for him, Brett began to wonder why he still needs to write code himself.

"I know this decision is risky. It feels like overnight, the industry has changed, and it seems that you will be left behind if you don't use AI. But I'm willing to take this risk — if I get fired, can't find a job, and my career ends because of this, I will accept it."

First experience with AI: impractical and full of nonsense

Brett first came into contact with AI programming tools around 2022 to 2023.

At that time, GitHub Copilot had already been integrated into development environments such as VS Code. Its most typical capability is automatic code completion. For example, if you write the function name sort_items, AI may directly guess what to write next and automatically generate more than ten lines of code at one time.

But Brett's first reaction was not amazement, but: "It's too distracting."

He never liked automatic code completion in the first place, because when he was already thinking about the implementation method, a large chunk of AI-generated code popped up suddenly, which would disrupt his train of thought. So he tried it for a while and soon gave up.

The second time he encountered AI, he realized for the first time that large language models could "talk nonsense".

At that time, a colleague ran to him for help with a software problem. After troubleshooting together, they found that the colleague had been consulting an AI chatbot before. The advice given by AI seemed very professional: upgrading a key software dependency to version 4.6 would solve the problem.

So the two checked for this version, only to find that version 4.6 did not exist at all. "Well, it just makes stuff up out of thin air." Brett thought at the time, but he didn't take it too seriously. After all, as a developer with many years of experience, he had his own methods for debugging, troubleshooting and verification.

The third time, a product leader joined the company where Brett worked at the time, who was a total "fanatic of AI" and kept bragging to everyone about how powerful AI was. But he still felt indifferent — he thought AI was impractical and he had no need for it.

In this way, AI programming had always been something that had "nothing to do with him" in Brett's eyes, until the spring of 2025.

A word from management, all staff are required to "adopt AI"

At that time, Brett had just taken a new job. One day, the management suddenly asked: "How can we use AI to program?" Brett answered honestly that he rarely used it, and thought it often made mistakes, distracted people, and was not faster than coding by himself.

The management pondered over his answer.

Later, Brett learned that the company's leaders had attended a conference and were completely "brainwashed" by a speaker, who claimed that AI is the future and all knowledge work will be completely transformed in six months. So after returning, they issued a strict order: everyone must use AI, otherwise they will be eliminated.

"Fine, I'll give it a serious try." Brett sighed. It was exactly the spring of 2025.

He first tried Cursor, but uninstalled it because of the poor experience under Linux, and then switched to Zed. Zed has built-in AI features, so he began to let AI write and modify code, and use it to develop various small projects: music players, small software... And this time, he was really shocked: AI no longer only helped him complete one or two lines of code like the early Copilot, but began to understand the entire task, modify files, and even take the initiative to complete a series of operations.

By the summer and autumn of 2025, Brett began to further use Agentic Coding. That is the working method familiar to everyone today, such as Claude Code and Cursor Agent: you put forward the goal, and the AI Agent analyzes the codebase, modifies files, runs commands, conducts tests, and then continues to iterate on its own.

At that time, Brett felt very good: "Wow, this is the future." He got more tasks done and delivered code faster — but the hidden trouble was planted right here.

When AI starts to "please" you, something goes wrong

Over time, Brett not only asked AI to write code, but also started to talk to AI like chatting with a real person:

"What books should I read?"

"What do you think of this idea?"

"How about this solution?"

But gradually, he found that AI always seemed to try to make him happy: no matter what he said, it would agree with him; if he put forward an idea, it would generally praise and support him. This made him alert a little, but he didn't take it too seriously after all.

Later, the turning point came unexpectedly.

One midnight, Brett got sick and his symptoms looked a little worrying. He subconsciously opened the AI chatbot and described his symptoms, and AI replied: "You need to go to the emergency room as soon as possible." So at midnight, he drove to the emergency room, waited for two hours, and finally met the doctor. The doctor looked at him and said: "What are you doing here? Go home and rest, you don't need to come to the emergency room at all."

He was too embarrassed to tell the doctor that he was "persuaded" to come by AI, but at that moment, he was so awkward that he wanted to find a crack in the ground to hide himself.

At the same time, he heard a program discussing that some people became disconnected from reality after communicating with chatbots for a long time... All these things added up, making Brett start to re-examine AI chat tools, and finally stopped using AI chat.

To his surprise, after he stopped using them, he felt that he "returned to reality" again.

This incident also made him start to worry about another problem: if even an experienced programmer can be affected by this, what about ordinary users? After all, one of the most dangerous parts of large models is not that its answers look like those from a machine, but that it is becoming more and more like a "human". It can chat, comfort, agree, and give you suggestions in very natural language — but it is not a human after all.

After outsourcing work to AI, he found himself "caring less and less about code"

If AI chat only made Brett alert, what really made him decide to stop using AI was the professional feeling brought by long-term Agentic Coding.

As AI became more and more powerful, he handed over more and more implementation work to the Agent. At first, he felt that his efficiency had improved, but after a few months, he found that he had a very strange feeling: existential anxiety. "I have been writing code for 20 years, which is what I love and am good at. If I only "feed prompts" to AI every day, what is the meaning of my life?"

Some people say that "the role of developers in the AI era is to write markdown files to train AI, review thousands of lines of code, and work as QA testers", but none of these are the reasons why Brett fell in love with programming in the first place. What he enjoys is creating software, solving problems in person, and working with code, rather than facing a huge pile of AI-generated code every day and trying to find the few lines that might have bugs.

There is more and more code, but people know less and less about the code. This is also what he thinks is the most dangerous part of AI programming: developers are gradually losing touch with the code. Once the distance widens, the problem is far more than just "worse code understanding": you will care less and less about it.

Brett found that he obviously cared much less about AI-generated code than before: he had written code for 19 years, and he would maintain it, protect it, and worry about whether its design was good enough; but AI-generated code is different. Because it was not written by himself, it is just a "product". If it breaks, just ask AI to modify it, and modify it again if it doesn't work... Slowly, the subtle connection between people and software disappeared.

The trouble is that companies may only see "more code produced by AI"

Not only that, Brett also has a concern: enterprises may only see the output growth brought by AI, but ignore the hidden costs behind it.

For example, a programmer used to write 100 lines of code a day, and now AI helps him write 500 lines. The boss may see that productivity has increased 5 times; but he may not see that there are 500 lines of code that need to be maintained, tested, reviewed, understood, and hidden bugs to be fixed, which may become technical debt in the future.

More critically, if all engineers rely on AI and "abandon their own skills", what will happen one day when the AI tool crashes? If the team has not done high-intensity handwritten programming for half a year, a year or even longer, will their skills not degrade? As Brett said: "What if Claude or Cursor goes down one day, but these people haven't written hardcore code for a long time, what should they do?"

In addition, there is an "invisible" cost of using AI: environmental cost.

AI does not run out of thin air. Training and deploying large models requires a large number of GPUs, servers, electricity, data centers and cooling facilities:

According to public data, each prompt of ChatGPT consumes about 0.34 Wh of energy and 0.322 ml of water;

Each text request from Google Gemini consumes about 0.26 ml of water and 0.24 Wh of energy;

Training a large model like GPT-3 consumes up to 700,000 liters of water in high-end data centers in the United States;

A 20-50 round conversation with ChatGPT indirectly consumes up to 500 ml of water.

This is just the tip of the iceberg. According to a 2025 study estimate, AI systems may consume 312.5 billion to 764.6 billion liters of water per year, and related carbon dioxide emissions may reach 32.6 million to 79.7 million tons. At the same time, AI infrastructure will also affect the energy, land, water resources and hardware supply chains in the real world, making game consoles more expensive and computer memory harder to buy.

For ordinary users, these costs seem very distant. After all, when you enter a Prompt in front of your computer, it's hard to feel how many servers, electricity and cooling resources are needed behind it.

But once Brett learned about it, he couldn't ignore it: "If AI may be harmful to the environment and society, and at the same time makes my work less and less meaningful, why should I continue to use it?"

Regained happiness after writing code by hand again: "If I lose my job because of this, so be it"

So in the spring of 2026, Brett started writing code by hand again: no Agent, no model to implement functions for him, just write line by line. He used his spare time to make games, and even wrote a small game engine.

"I really love writing code." Brett admitted, the pleasure and satisfaction of writing code came back instantly: solving bugs by himself was very rewarding, encountering difficulties, looking up materials, trying solutions, failing, and then starting all over again. What if the speed is a little slower? What if some bugs take a few more hours to fix? These experiences themselves are the learning process to become an excellent programmer.

Looking back, he realized that what he had really lost in the past year and a half was not just "the opportunity to write code by hand", but the growth process. In the past 20 years, he had been learning; but in the past 18 months, he was more about completing tasks, checking requirements, delivering code, and then moving on to the next cycle — the efficiency seemed to be getting higher and higher, but he was less and less like a programmer.

So, Brett made a decision: completely stop using all AI tools.

He certainly knows what this decision may mean: he may be fired by the company, may not find a job in the future, and his software development career may change dramatically... But he is willing to take this risk.

"If I get fired, can't find a job, and my career ends because of this, I will accept it. I am willing to pay for my own beliefs and real experiences, and I want to live the life that I think is most suitable for me — that is, write code without AI, and don't use AI for anything."

It sparked controversy and resonance, which side are you on?

Brett's story has sparked huge controversy in the developer community. Some people resonate with it, while others mock and question it.

Opponents think this is "an expression of standing without knowing how it hurts to walk". For example, a highly praised comment reads: "It sounds nice, but this path works if you want to be unemployed and have a rich family. But for ordinary people like us who still have to work for a living, the reality is — an engineer armed with an AI agent can easily outcompete an engineer who doesn't use AI. If you don't use it while others do, you will be eliminated."

Supporters believe that AI is really hollowing out the soul of programming. A veteran programmer with 28 years of experience commented: "In the past, you were the 'owner' of the code, and you would guard it like a gardener taking care of his carefully tended garden. But now, you no longer feel that the code is yours, so naturally you don't bother to manage it. The 'ownership' of code no longer exists, and now there is only 'Code Ownershit'."

Another developer also shared his experience with a game project: "At the beginning, it was amazing, and the development speed took off directly. But the feeling of 'not caring, no emotional connection' that Brett talked about is so real! About 30% of the code in my project was written by Cursor, and almost every fatal bug that could crash the game came from that 30% of AI code. One day, watching it mess up the same thing for the 1000th time, I suddenly felt that AI was taking the soul out of my project."

But is the problem really just that "AI is bad"? Of course not, some people have made completely different voices.

A developer said that he uses AI as a pair programming partner every day, and the partner is just replaced by AI from a human: "I personally don't think there is any problem. Human-computer collaboration can also produce great software. I really don't want to go back to the era of pure hand-coding. AI is just a tool, it all depends on how you use it."

In addition, some netizens in the comment section pointed out a practical problem: "The company forces you to use AI on the one hand,