Use It or Lose It: Developers Confront the Hidden Costs of AI Coding Tools

AI programming tools (such as Claude Code and various large language models) are reshaping the software development process, rekindling the enthusiasm of senior programmers, and at the same time, triggering widespread concerns about code quality.

All this may have started in February. At that time, 51-year-old programmer and entrepreneur Paul Ford wrote an article for The New York Times, enthusiastically declaring that "the era of AI disruption has arrived." AI programming tools are becoming more and more powerful, and the hopes and fears of programmers who may be replaced by them are also intensifying simultaneously. A chaotic and intense discussion has emerged on the Internet.

In the article, Ford praised the possibilities brought by the faster and cheaper "Vibe Coding." While revisiting those long - shelved projects, he wrote, "It's really interesting to see old ideas come to life."

By mid - March, technology writer Clive Thompson published an interview in The New York Times with more than 70 software developers from Google, Amazon, Microsoft, and several startups. The interviewees generally said, "Most people are genuinely excited about their new abilities. To my surprise, so many developers admitted that they were glad they didn't have to write code by hand anymore. Many people mentioned that they still felt the joy of success even when the code was generated by AI."

But some people expressed deeper concerns. Pia Torain, a software engineer at Point Health AI, told Thompson that after sending hundreds of prompts every day for four consecutive months, she "began to lose her programming ability." Now, Torain deliberately slows down and carefully understands the entire program's architecture and process. She warned that outsourcing too much programming work to AI may ultimately bring great risks.

"Use it or lose it."

"Use it or lose it," Torain told Thompson. This puts junior developers in a more difficult situation, as summarized by a comment on Hacker News: "If large language models can complete junior - level work faster and cheaper, resulting in no one willing to hire junior developers, then who will grow into experts in the future?"

This kind of concern is manifested in various ways. A Hacker News commenter sighed, "I used to think about teaching when I got older. But now students can find countless teaching resources on YouTube, and now there are Gemini, Claude, ChatGPT. These tools are all too good."

They already feel that as human teachers, they seem to have nothing to contribute, and with the rapid progress of the tools, this value may completely disappear in two years.

The improvement in coding efficiency also brings corresponding problems: it not only raises concerns about code quality but also makes people think about whether professional knowledge will be more valuable or gradually depreciate in an era filled with powerful AI programming assistants. At the same time, it also increases the risk of developer burnout.

As new tools continuously change the industry ecosystem, relevant discussions are becoming more and more intense. Top - tier media and online communities are recording this transformation from multiple perspectives, showing that the entire industry is facing the impact and transformation brought by AI programming tools - including both positive and negative consequences.

Enthusiasm and Speed vs. Quality and Professional Knowledge

Earlier this month, a 60 - year - old programmer posted on Hacker News that Claude Code "rekindled his enthusiasm," and a related discussion immediately broke out. The post received 1086 likes and 989 comments, including more relevant experiences and insights, as well as controversies and concerns. (A few days later, a post with an almost refuting stance appeared, saying that Claude Code "is killing enthusiasm.")

In the past few months, more powerful AI tools have significantly improved the work efficiency of some programmers, especially those in the later stages of their careers. Among these 989 comments, a programmer named John Calhoun, who claims to be an old - school Macintosh shareware developer and joined Apple in 1995, admitted that "he created a website that he wouldn't have otherwise attempted through Vibe Coding... and also put several long - shelved projects back on the agenda..."

The phenomenon of long - shelved projects coming back to life is happening all over the world. In Dresden, 62 - year - old programmer Reini Urban said that the large language model coding assistant "rekindled my enthusiasm. Those old projects that were previously stalled due to high difficulty will soon be restarted. With Opus, all this has finally become possible." And 51 - year - old Boston electrical engineer and entrepreneur John Reine posted, "It gave me the courage to become an independent founder."

But among the 70 software developers interviewed by Thompson, a small number of people clearly "actively avoided" AI tools. One of them is an Apple engineer who prefers to remain anonymous - he still insists on doing all the coding himself. "I believe that programming itself can be interesting and fulfilling, and letting the computer do it for you will deprive you of these experiences," this engineer said, expressing his reluctance to "outsource" his enthusiasm.

There are also concerns about code quality. As a Hacker News commenter said, "Large language models are good at writing code, but they perform poorly in software engineering... Currently, I'm fixing an application written through Vibe Coding. Each individual function seems okay, but the whole thing is a mess like a plate of spaghetti, causing a lot of problems."

One of the commenters is Joel Dare, a 50 - year - old software engineer from Utah. He sighed that after working in the industry for 40 years, "my tolerance for architectural degradation has become extremely low." Once, he forgot to prompt Claude not to use a framework in a Node project, and the generated project actually had 89 dependencies. "In this industry where we prioritize'speed' over maintainability, this situation has become the norm. But for me, this is completely unacceptable. I'll try using AI tools again, but for now, we at least need to be able to skillfully and reasonably control these tools..."

Many commenters said that in the final analysis, AI tools make their professional knowledge more valuable. A chief software engineer believes that AI "amplifies what I'm good at" - including architectural design, problem debugging, and "making high - quality technical decisions."

63 - year - old Chris Marshall, who has been programming since 1983, said, "I really think that my experience is crucial for developing deliverable products with the help of large language models. I have to learn how to collaborate with large language models, and I think I've found my rhythm."

Software engineer Juan Reyero has a unique analysis. With 40 years of programming experience, he believes that this debate is gradually dividing into two camps. "In my opinion, the most core difference is that one group of people enjoys programming itself - carefully crafting beautiful code... while the other group values more that the code can be completed, has a clear structure, and runs normally. They often regard the process of writing code as an annoying intermediate step."

As Dare said, AI coding "still requires our professional knowledge to guide it."

But he also added, "I'm not sure if it will still be the same in a year, but it is the case for now."

Good News and Bad News

Perhaps 57 - year - old software engineer Steve Yegge best represents people's complex attitude towards AI programming tools. In January this year, on the verge of his 57th birthday, he posted an article on Medium, sighing that he can now "write thousands of lines of production - level code every day... and generally enjoy it," to commemorate this new - spreading enthusiasm. (In The New York Times article, Yegge said that his current productivity is 10 times, 20 times, "even 100 times" that of other periods in his career, "It's like we've been walking on our legs all our lives, and now we're suddenly flying.")

Yegge labeled his occupation on his LinkedIn profile as "AI nanny," and he is enjoying the exponentially improved work ability brought by powerful AI agents and orchestration tools.

But for Yegge, this explosive increase in productivity has also brought unexpected side effects. Yegge and his two colleagues admitted, "High - intensity Vibe Coding has disrupted our sleep rhythm." ("This didn't happen last year until we started connecting to more than a dozen or even more AI agents simultaneously and clustering a large number of tasks...")

In the Gas Town project, an AI agent platform launched by his startup, "Our idea was to only make high - level decisions, but our brains were working too hard, and we were completely exhausted. We had to take a break before we could continue." But "for me, this state makes me feel physically and mentally exhausted."

A Boon for Senior Programmers?

Perhaps this new ability is particularly meaningful for senior programmers. 64 - year - old programming master Kent Beck even told Thompson that he basically stopped writing code ten years ago and was once disappointed with programming languages and tools, but large language models made him pick up programming again. (Beck admitted that this kind of fun is very addictive and said that the unpredictability of AI "is as addictive as a slot machine.")

63 - year - old Chris Marshall even believes that AI solves a hidden age - discrimination problem. "The most frustrating thing after retirement is that no one wants to work with me," Marshall posted in an online discussion. "I've worked in teams throughout my career, and being forced to work alone greatly limits my performance. Large language models have given me the courage to dream again." Although he is still a retired programmer, he wrote, "I'm very happy to have a large language model as a pair - programming partner."

As the comments on Hacker News increased, some senior programmers even detailed the reasons that attracted them to use AI - assisted programming. 52 - year - old Douglas Tarr wrote that "purely manual" coding made him recall the days when he worked 12 hours a day. "Just thinking about it makes me tired... I'm older now, sitting for too long gives me back pain, and occasionally looking at the screen for a long time gives me migraines..."

And the most touching response came from programmer TQ White in Minnesota. He posted that he wrote his first program as early as 1967, and AI coding tools solved his biggest problem - "the loneliness of being a retired programmer is really a big problem..."

White joked, "I don't dare to act like I think AI is a good enough substitute, or those who criticize me will say I have an unhealthy mindset..." But then he changed the subject: "Between you and me, although I miss the friendship of working with real people, it's also quite nice to brainstorm with a partner who knows almost everything and can execute according to my ideas without complaint."

"Moreover, it's really great to have someone or something to discuss technical ideas with. It's really great to live in this era..."

Original Link: https://thenewstack.io/ai-coding-tools-reckoning/

This article is from the WeChat official account "AI Frontline," written by David Cassel and published by 36Kr with authorization.