80% 的开发者认为 AI 编程更令人上瘾，而非更有帮助

Tomasz Śmigla / iStock / Getty Images Plus via Getty Images

ZDNET 核心要点

• 程序员认为 AI 工具令人上瘾，但也令人疲惫。
• 修复 AI 编写代码所造成的延迟，可能抵消编程速度的提升。
• AI 编程工具可能导致程序员陷入工作成瘾和职业倦怠。

是的，是的，我们明白。得益于 GitHub Copilot、Claude Code、Cursor，以及任何当下火热的新型开发者 AI 工具，程序员能完成的工作比以往更多。但即使开发者越来越多地使用这些工具来生成样板代码、解释陌生代码、起草测试、重构模块和排查错误，许多人也发现，这些工具同样可能相当令人头疼。

另请阅读：用 AI 替代员工为何会适得其反，以及聪明的领导者创造真正价值的 5 种方法

AI 引发的工作成瘾

AI 驱动的事故报告公司 Rootly 的 CTO 兼联合创始人 Quentin Rousseau 曾在一篇 LinkedIn 帖子中这样描述：“现在是凌晨 2 点 47 分……我不是在排查故障，也没有截止日期。我只是看着 Claude Code 重构一个模块……而且停不下来。”为什么？他接着说：“Agentic coding 令人上瘾。当 Agent 把事情做对时，你会获得多巴胺刺激；当它失败时，你又会感到肾上腺素飙升。”Rousseau 承认自己无法入睡，并不得不寻求医疗帮助。

这并不太令人意外。程序员长期以来就容易陷入工作成瘾。不过，AI 为工作带来了新的节奏。正如 Rousseau 所说：“看着一个 Agent 工作，既足够被动，让人觉得像是在休息；又足够主动，让你持续沉迷其中。”结果便是一种新型的职业倦怠。

另请阅读：这位 AI 专家称，即使你是程序员，岗位末日也不会到来——原因如下

这并不只是某一位程序员的经历。AI 编程可能把软件开发变成一个永不停歇的反馈循环。开发者不再是完成一项任务后就离开，而是可以不断要求 Agent 提供另一种实现、重写、优化或重构方案；与此同时，他们还会被一种挥之不去的担忧困扰：停止工作就意味着还有任务未完成。

当 AI 从助手变成枷锁

编程培训公司 Coddy Tech 对 305 名开发者开展的调查发现，“五分之四的开发者（80%）表示，他们的AI 使用体验更像是一种依赖，而非一种优势。”当然，受访者认为这些工具有用，但他们也担心，习惯性依赖 AI 会削弱自身的问题解决能力、扩大工作量，并助长一种不健康的工作关系。

因此，“超过五分之二的开发者（43%）即使原本打算停止，也会在下班后继续使用 AI 编程；32% 的人曾推迟睡眠以继续工作。”此外，39% 的受访者表示，AI 工具让他们更难从工作中抽离。

另请阅读：我调查的开发者中，75% 更偏好 Claude Code——他们为何选择它而非 Codex

问题不只在于 AI 工具的成瘾性。74% 的开发者报告称，重度使用 AI 会增加他们获得加薪或晋升的可能性。然而，51% 的人也表示，这会增加他们职业倦怠的可能性。

“信任，但要验证”：AI 的烦恼

正如 2025 Stack Overflow 开发者调查所发现的那样，45% 的受访者会因 AI 给出“几乎正确、但并不完全正确”的答案而感到沮丧。结果就是：输出看似很有说服力，却制造了棘手的调试工作。

Stack Overflow 的研究还发现，尽管 AI 工具的采用率仍在上升——目前已有 80% 的开发者在工作流中使用这些工具——但对 AI 准确性的信任度已从往年的 40% 降至今年的仅 29%。因此，程序员对 AI 的正面评价同比从 72% 降至 60%。

另请阅读：一项来自过去的低技术解决方案，或许是你防御 AI 深度伪造的最佳手段

易用性与更重工作负担之间的摩擦至关重要。除了要弄清 AI 到底哪些地方“几乎做对了”之外，开发者仍必须理解需求、识别生成代码何时会与系统架构冲突、测试边缘情况、处理安全风险，并对生产环境后果负责。

这会形成“验证债务”（verification debt）。输出很快就会送达，但你仍需要确认它是否正确、安全、可维护，以及是否适合特定代码库。

此外，Coddy 调查所描述的依赖现象，可能会因雇主如何解读 AI 驱动的产出而进一步被放大。如果一家组织将 AI 视为成倍提升开发者产能的方式，员工就可能面临在相同工作时长内交付更多功能、关闭更多工单、完成更多审查的压力。

这反过来会抹去单项编程任务节省下来的时间，并将负担转移到其他地方：更大的拉取请求（pull request）、更多需要检查的生成式变更、更多需要验证的依赖项，以及更多需要管理的运营风险。

另请阅读：你可以免费获得 Google AI Professional Certificate，我强烈推荐

因此，AI 辅助编程正逐渐成为一个与工作生活平衡同样相关的问题，而不只是一个工具问题。利用 Agent 消除日常琐碎工作的团队，可能会看到实实在在的收益；但试图利用它加速软件生产管道每个环节的团队，则有可能创造出同一份工作更快、更无情的版本。

对于 Rousseau 这样的开发者而言，担忧已不再仅仅是 AI 能否编写代码。它能做到，我们明白。接下来真正的问题是：开发者是否仍能决定工作日以及与 Agent 的循环在何时结束。

Tomasz Śmigla/ iStock / Getty Images Plus via Getty Images

Follow ZDNET: Add us as a preferred source _on Google.
_

ZDNET's key takeaways

• Programmers find AI tools addictive, but exhausting.
• Coding speed can be lost due to delays in fixing AI-written code.
• AI programming tools can lead to programmer workaholism and burnout.

Yes, yes, we get it. Thanks to GitHub Copilot, Claude Code, Cursor, whatever the hot new AI tool is for developers, you can get more work done than ever before as a programmer. But even as developers use them ever more to generate boilerplate, explain unfamiliar code, draft tests, refactor modules, and troubleshoot errors, many are also finding that the same tools can be pains in the rump.

Also:Why replacing staff with AI backfires - and 5 ways smart leaders generate real value instead

AI-induced workaholism

That's because, as Quentin Rousseau, CTO and co-founder of the AI-powered incident report company Rootly, described in a LinkedIn post, "It's 2:47 a.m... I'm not debugging an outage. There's no deadline. I'm just watching Claude Code refactor a module… and I can't stop." Why? Because, he continued, "Agentic coding is addictive. When the agent gets things right, you get a dopamine hit. When it fails, you get an adrenaline rush." Rousseau confessed that he couldn't sleep and had to seek medical help.

This isn't too surprising. Programmers have long been prone to workaholism. However, AI has brought a new pace to the work where, as Rousseau put it, "Watching an agent's work is passive enough to feel like rest, active enough to keep you hooked." The result is a new kind of burnout.

Also:This AI expert says the job apocalypse isn't coming, even if you're a coder - here's why

This isn't just one programmer's experience. AI coding can turn software development into an always-on feedback loop. Rather than completing a task and stepping away, developers can continually ask an agent for another implementation, rewrite, optimization, or refactor, or combine it with the nagging worry that stopping means leaving work undone.

When AI becomes a ball-and-chain instead of a helper

Coddy Tech, a programming training company, found in a survey of 305 developers that "Four in five developers (80%) say their AI use has felt more like dependence than an advantage." Sure, they found these tools useful, but they also worry that habitual reliance on AI is weakening their own problem-solving process, expanding their workload, and fostering an unhealthy relationship with work.

So it is that "more than two-fifths of developers (43%) keep coding with AI after hours even when they meant to stop, and 32% have put off sleep to keep going." Also, 39% said AI tools have made it harder to switch off from work.

Also:75% of developers I surveyed prefer Claude Code - here's why they choose it over Codex

It's not just the addictive nature of AI tools. Heavy use of AI, reported 74% of developers, made it more likely they'd earn a raise or promotion. However, 51% also said they were more likely to burn out.

Trust but verify AI annoyances

It also doesn't help any that, as the 2025 Stack Overflow Developer Survey found, 45% of respondents were frustrated by AI answers that were "almost right, but not quite. The result? Output that appears convincing while creating difficult debugging work.

The Stack Overflow study also found that while AI tool adoption has continued to climb, with 80% of developers now using them in their workflows, trust in AI accuracy has fallen from 40% in previous years to just 29% this year. As a result, programmers' positive favorability toward AI has decreased from 72% to 60% year over year.

Also:A low-tech solution from the past may be your best defense against AI deepfakes

The friction between ease of use and a heavier workload matters. Besides trying to work out what the AI almost got right, developers must still understand the requirements, recognize when generated code conflicts with a system's architecture, test edge cases, address security risks, and own the production consequences.

This creates "verification debt." The output arrives quickly, but you're still stuck establishing whether it is correct, secure, maintainable, and appropriate to the specific codebase.

On top of that, the dependence described in the Coddy survey may be amplified by how employers interpret AI-driven output. If an organization treats AI as a way to multiply developer capacity, workers can face pressure to ship more features, close more tickets, and perform more reviews in the same number of hours.

That, in turn, can erase the time saved on individual coding tasks and shift the burden elsewhere: Larger pull requests, more generated changes to inspect, more dependencies to validate, and more operational risk to manage.

Also:You can earn your Google Al Professional Certificate for free - and I highly recommend it

AI-assisted programming is, therefore, becoming as much a life-work balance issue as a tooling issue. Teams that use agents to remove routine toil may see genuine benefits. Teams that use them to accelerate every part of the software-production pipeline risk creating a faster, more relentless version of the same job.

For developers like Rousseau, the concern is no longer merely whether AI can write code. It can. We get that. The question going forward is whether developers can still decide when the workday and the agent loop end.