突然发现已经很久没在 juejin 平台静下心来写一篇技术博客了，因为我现在写的博客介绍的技术都是从 AI 来的，总是让我有一种不是我自己真本事的感觉，AI 时代也许死磕技术真的走到头了。
下面介绍一些我对使用 AI 这一年多的直接感受，我称之为 AI 副作用。
AI 让我丧失了深度思考的能力
这话听着有点反直觉——AI 学会了深度思考，我却学不会了。
以前遇到线上那种难缠的组件故障，我的流程是固定的：翻日志、追源码、画调用链，一坐就是半天，最后揪出那个藏在犄角旮旯里的 bug，那种成就感是真的爽。现在呢？把问题描述往 AI 那一递，我往旁边一靠，拿起手机刷一会儿，等着答案送上门。
答案来得是快，可我发现自己的"故障肌肉"在退化。以前那种顺藤摸瓜的直觉，是需要一次次亲手扒源码喂出来的，现在这条路径被 AI 直接截断了。
更明显的是对技术的态度。新名词好像每天都能冒出来一批——Harness、Agent、Skill……我常常是看个概念、跑个 demo 就算"了解"了，浅尝辄止。久而久之，我怀疑自己正在变成一个半调子程序员：什么都知道一点，什么都不深。
还有个挺有意思的观察。公司内部技术论坛里，纯分享一个知识点的博客越来越少，大家都往 AI 方向写。无他——你认真介绍一个普通知识点，写得还不一定有 AI 好；而写 AI 方向的最新进展，哪怕内容肤浅，流量也管够。这大概是市场给出的最诚实的信号。
AI 让我陷入了自我否定
我 18 年左右入行，那时候还是"技术为上"的年代。刚入行的我能为了研究 JVM 原理死磕，买一堆书，周末还泡在网上看教程。那种学习在没有 AI 的时代，是会一点点沉淀成我的技术资产的——你学过的每一样东西，都长在自己身上。
可 AI 来了之后，这套逻辑就不太成立了。
以前学新东西能让我有一种"比别人强"的感觉，现在这种感觉很难维持了。有 AI 在场，我很难说自己比一个刚入行不久的程序员强多少——你知识面再广，也广不过一个随时在线、读过整个互联网的模型。
写博客的心态也变了。一开始那种"终于弄懂一个问题、迫不及待要分享"的快感，现在很难找回来。这感觉就像打游戏时穿着一身神装去砍瓜切菜，三两下把 boss 打倒、快速通关——爽是爽了，但留给你的只有虚无。你赢的不是游戏，是装备。
网上现在到处在宣扬"谁都能自己写代码"。说实话，看到这些，各位程序员心里会不会有一丝不甘？毕竟自己攒了这么多年的技术，AI 一来，这道门槛就被踏得平平整整。
AI 让程序员这个职业确实变得岌岌可危
这一点我觉得大多数人不会否认。虽然现在我们还像是"骑马的人"，还能起到一点指挥的作用，但随着 Agent 越来越完善、大模型底座能力逐步增强，这匹马大概率会朝着"不需要马夫"的方向跑。到那时候，缰绳还在不在你手里，是另一回事。
AI 时代看起来机会很多，其实普通人的机会很少
现在有一种很流行的论调：普通人只要想出一个很棒的 idea，用 AI 做出一个惊艳的作品，就能财富自由、大赚一笔。
我觉得这想法挺天真的。AI 时代就算你真做出一个亮点，这个东西本身是没有技术门槛的——任何人都能直接用 AI 再做一个跟你一样的。你的护城河不是那个作品，而是别人暂时还没想到。可"没想到"这件事，在信息传播这么快的年代，保质期短得可怜。
我时常觉得，我们用 AI，就像当年美洲的淘金热——最终大赚特赚的，从来不是淘金的人，而是卖铲子的那些。
写完这些，我倒也不是想唱衰什么。AI 是真有用，我也天天在用。只是作为一个在行业里待了快十年的人，我觉得有些变化值得被记录下来——不是结论，就是一个普通程序员的心里话。如果你也有类似的感觉，那说明我不是一个人。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
And suddenly it's been a long time since I wrote a technical blog at the juejin platform, because the technology I'm writing now comes from AI, and it always gives me the feeling that I'm not really capable of doing it myself, and that the age of AI may be over.
Here are some of my direct feelings about using AI for over a year, which I call an AI side effect.
AI lost my ability to think deep.
That sounds a little counterintuitive -- AI learned to think deep, but I can't.
I've had some kind of entanglement failure on the line, and my processes are fixed: flipping the logs, tracking the source code, painting the call chain, sitting for half a day, and finally pulling out the bug that's hidden in the corner, and the sense of achievement is really good. And now? Describe the question to AI, and I lean to the side, pick up my phone and brush it, and wait for the answer to the door.
The answer came quickly, but I found my own "failed muscles" deteriorating. It used to be the feeling that it had to be fed with the source code once and for all, and now it's been cut directly by AI.
Even more obvious is the attitude towards technology. It's like every day new names come out -- Harness, Agent, Skill -- and I'm always looking at a concept, running away, even if I know it, it's easy. For a long time, I suspected I was becoming a half-twisted programmer: I know everything, I know nothing.
There's another interesting observation. In-house technology forums, there are fewer and fewer blogs that share a single point of knowledge, and everyone writes in the direction of AI. Without him -- you don't have to have an AI in mind; you can write the latest developments in the direction of the AI, even if it's shallow, it's got enough traffic. This is probably the most honest signal from the market.
AI got me into self-denial.
I started in about 18 years, and then it was the "technology first." I just got into business and I bought a bunch of books to study JVM principles, and I was watching classes on the Internet on weekends. That kind of learning without an AI is a little bit of my technical asset -- everything you've learned has grown on you.
But when AI came, the logic didn't work out.
I used to learn something new to make me feel better than others, and it's hard to keep that feeling. It's hard to say how much better I am than a programmer who's just started his career — you know a lot, and you can't read the whole Internet on the Internet at all times.
The mentality of blogging has changed. At first, the pleasure of " finally getting to know a problem and dying to share it" was hard to find. It's like playing a game in a costume to cut the guacamole, beating the boss down in three or two, breaking off customs fast -- it's cool, but nothing left for you. You won not games, you won equipment.
It's all over the Internet now, "Everybody can write their own code." To tell you the truth, do you guys have a problem with this? After all these years of technology, AI came and the threshold was even.
AI, making the job of a programmer really dangerous.
I don't think most people would deny that. Now we're like a horse rider, and we can do a little command, but as Agent gets better and LLM gets better, the horse probably runs in the direction of "no needing a horseman." It's a different matter if you still have the rope.
AI, the times seem to be a lot of opportunities.
Now there's a very popular argument: ordinary people can make a lot of money and make a lot of money if they think about a great idea, make an amazing work with AI.
I think it's naive. Even if you do make a bright spot, there's no technical threshold for this thing -- anyone can use AI and make another like you. Your moat isn't that work, but it's something others haven't thought of yet. But the "unexpected" thing, in the time when information was spreading so fast, the quality period was so short.
I always think that we use the al, like the gold rush of the Americas -- the ones that end up making a lot of money, not the gold hunters, but the shovels.
I don't want to sing shit after all this. AI is really useful, and I use it every day. Just as a person who has been in the industry for almost 10 years, I feel that there are changes worth recording — not conclusions, just the words of an ordinary programmer. If you feel the same way, it means I'm not alone.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.