AI 会挤掉新人吗？会。所以有实力的人，欠新人一样东西
先讲答案：AI 会结构性地挤压新人的空间。这已经不是感觉，薪资数据看得到。
Stanford 数位经济实验室翻了美国几百万笔薪资纪录（ADP；论文〈Canaries in the Coal Mine?〉2025 年 11 月版）：最容易被 AI 影响的那些工作，从 2022 年底开始，22 到 25 岁的人，就业比影响小的工作少了 16%。同一批工作里，30 岁以上的人反而还在增加。
同样的职位，工作没有消失，消失的是入口。
原因不难懂。以前交给新人做的事：第一版稿、第一刀剪、整理表格、粗分镜，刚好就是 AI agent 几分钟、几毛钱就做完的事。你已经知道好是什么样子，就不需要有人先帮你做一版；你需要的是有人判断。那个人就是高手。
高手会越来越值钱，前提是拥抱 AI
2016 年，Hinton 说别再训练放射科医生了，五年内 AI 看片会比人强。现在放射科到处都是 AI，Mayo Clinic 的放射科医生却从 2016 年大约 260 人，变成 400 多人（TechCrunch，2025 年 5 月）。看片是任务，判断病人哪里出了问题才是目的，後面这一块反而更值钱。
前提是你真的在用。不碰 AI 的高手，经验保护不了你；在用的高手，一个人做一个小团队的事。
然後是我最想讲的
如果你就是那个高手，最省事的做法是以後都不请新人了，反正你一个人也做得完。
可是你认识的每一个高手，包括你自己，都是有人信过他、交给他一件他会做错的事，才长出来的。
所以当你有足够的实力容错的时候，就给新一代人犯错和成长的空间和机会：给他一件真的工作，让他交回来不完美，再跟他讲为什么。AI 可以写第一版，可是它给不了任何人「做错很多年」这件事，而新人就是在那几年里，变成知道好是什么样子的人。
现在没人有机会犯错，十年後就没有高手可以请。
Is AI taking entry-level jobs? Yes. Here's what the people who stay owe the ones coming in
Short answer: AI is squeezing the room for people who are just starting out. It isn't a vibe anymore. It shows up in payroll data.
Stanford's Digital Economy Lab went through millions of US payroll records from ADP ("Canaries in the Coal Mine?", November 2025 version). In the jobs most exposed to AI, employment for 22 to 25 year olds fell 16% compared with less exposed jobs, starting from late 2022. In those same jobs, workers aged 30 and over kept growing.
Same job titles. The work didn't disappear. The way in did.
It's not hard to see why. The tasks we used to hand to someone new (the first draft, the first cut, cleaning up a spreadsheet, a rough storyboard) are exactly what an AI agent now does in minutes for a few cents. If you already know what good looks like, you don't need someone to make a first version for you. You need someone to judge it. That's the senior.
Experts get more valuable. On one condition.
In 2016, Geoffrey Hinton said we should stop training radiologists, because AI would read scans better within five years. AI is now all over radiology. Mayo Clinic's radiologists went from about 260 in 2016 to more than 400 (TechCrunch, May 2025). Reading the scan was the task. Working out what is wrong with the patient was the point, and that part became worth more, not less.
The condition: you actually use the thing. Experience doesn't protect a senior who refuses AI. A senior who uses it does the work of a small team.
Now the part I believe most.
If you are that senior, the easiest move is to never hire a junior again. You can get it all done without one.
But every senior you know, you included, got there because someone trusted them with work they were going to get wrong.
So once you're strong enough to absorb a mistake, absorb one. Give a new person real work, let it come back imperfect, and walk them through why. AI can write the first draft. It can't give anyone the years of getting things wrong that turn a beginner into the person who knows what good looks like.
If nobody gets to make mistakes now, there will be nobody to hire as a senior ten years from now.