An accountant walks past a fruit shop on his way home. The owner's stressing out. Says he can't write social media posts to save his life. Foot traffic's drying up.

The accountant goes home, opens AI, writes a few posts, sends them over. Owner's happy, slips him some cash. Comes back for more later.

Here's my question. Tons of people open AI. Some produce stuff nobody reads. Some produce stuff that sells.

What's the difference?

AI is an amplifier. An amplifier amplifies whatever you feed it.

Feed it judgment, it amplifies judgment. What gets amplified is the judgment already sitting in your head.

There are roughly three kinds of brains.

One has real accumulated experience. Years in an industry. Instincts, taste, judgment. AI amplifies that into something valuable.

Another is half-full. Knows a little, not quite enough. AI amplifies it into a rough draft. Usable. Nothing special.

Then there's the empty brain. AI amplifies that into a pile of correct but useless nonsense.

Since we're talking amplifiers, let me draw you a map. What does AI actually amplify well?

What it's good at. Drafting, organizing information, formatting, data analysis. Sixty percent of the grunt work—it takes that over. Does it pretty well.

What it struggles with. Things that need original judgment. Emotional depth. Ask it to write a song—structure's there, lyrics rhyme. You listen. Something's off. Can't put your finger on it.

What it can't do. Real value judgments. Aesthetic choices. Building trust between people. It can't touch these.

That's about forty percent. Forty percent nobody can do for you.

AI is good at combining existing knowledge fast. Every output is basically a statistical recombination of what's already out there. The song it writes you—lyrics are fine, rhythm's there. You feel something's missing. That missing piece? You have to fill it in yourself.

It takes over the drafting grunt work. Saves your time. But figuring out what to write in the first place—it can't do that. That's on you.

Back to that accountant.

His useful posts had nothing to do with knowing how to use AI. Plenty of people know that.

What did he have? Judgment about what the neighborhood customers like. He knew which fruit to push, how to pitch it. Knew how to sell yangmei this season.

That stuff was already in his head. AI just turned it into publishable copy.

Send in someone with an empty brain instead. AI generates posts too. Nobody buys. Same tool, different result. The difference is what's in the head.

I was scrolling short videos till two in the morning last night. It kept pushing topics I'd already watched. Got more boring by the minute. Turned off the phone, lay there, couldn't sleep.

Has nothing to do with what we're talking about.

Let me tell you about a real person. After you hear his story, you'll understand what AI actually amplifies.

“In Sydney Making Peace”—that's his handle. Thirties, lives in Australia. Builds SaaS products.

He started doing indie dev on the side in September 2023. Quit his job in June 2024 to go full-time. Two partners, no employer-employee thing, just a team.

Twenty-four independent products. Three or four are the money-makers. A hundred thousand dollars a month in revenue.

In the last thirty days, he burned through sixty-five billion tokens. In two weeks, he shipped major updates to six products, submitted hundreds of commits—all AI-generated.

He built thirteen AI agents. One picks topics, one writes, one edits, one formats, one publishes. Fully automated. He doesn't touch the pipeline.

Customer acquisition runs on AI too. Four channels.

SEO—AI generates content, mines long-tail keywords, analyzes competitor traffic. Ads—AI connects to ad platforms, adjusts hourly. Influencer marketing—AI finds influencers, writes outreach emails. Organic traffic—AI runs a Twitter account matrix.

He publishes ten tweets at once. Averages a hundred and ten thousand clicks per post. He didn't change a word.

He works nearly a hundred hours a week. Most of the work is AI doing its thing. Meanwhile he's doing a PhD, doing business consulting. His biggest fear—internet goes down, or AI compute fails.

Notice how he changed.

Early on, AI was just a fancy search engine for him. He'd ask a question, AI gave an answer. He was using AI.

Then something shifted. He moved into human-AI collaboration. Stopped assigning specific tasks. Just gave macro direction, goals, KPIs. Became the boss of AI.

Think about that shift.

Before, he asked AI questions. AI was a tool. After, he gave AI direction. AI was a subordinate. He was amplifying his own judgment.

He had stuff in his head—product sense, direction, market understanding. That alone could've produced a few products. AI walks in, and suddenly it's not a few products. It's twenty-four. Thirteen agents. A hundred grand a month.

Same brain. Ten times the output.

段落图：程序员爸爸像老板一样指挥多个AI小助手在流水线上协作，体现一人大脑借助AI放大十倍产出

So if judgment is the bottleneck, what's next?

Stop wasting time learning more AI features. You'll never learn them all. Most of what you learn won't help anyway. You need to figure out one thing—what should I have AI do?

Here's a method. Stop dumping everything on AI. Let it expose your blind spots.

How? Ask it a question. If you can't understand its answer, or you can't tell if the answer is good—that's your blind spot.

Once you see it, go study that area.

After you fill it in, your judgment grows. Next time you command AI more effectively. It exposes a new blind spot, you fill that one too.

Round and round. A positive loop.

Flip it around. What if you don't fill the gaps? What if you just keep using?

You dump everything on AI, learn nothing yourself. Your judgment doesn't grow. It shrinks. You lose the ability to tell if AI's output is any good. Eventually you can't even answer what AI should do for you.

There was a company in Guangzhou selling AI training. Called Qingzhou Intelligence. Got shut down by police in January. Fraud charges. Their victims? People who thought learning AI would make them money, but had nothing in their heads.

Two MIT PhD students ran an experiment. Published in Science. Four hundred fifty-three professionals doing writing tasks. Half used ChatGPT.

The ones using it finished forty percent faster. Quality went up eighteen percent. Here's the interesting part—AI mainly replaced the drafting stage. Idea generation and editorial judgment became the new bottlenecks. Judgment got more valuable.

That forty percent of time saved? It was all the brainless stuff. Typing, formatting, checking references.

Don't celebrate too soon. The work that's left? Every single piece requires you to make a call. Take a stand. AI can't do it for you.

That judgment is your livelihood.

I use AI to write too.

One time it handed me a paragraph. Smooth read. I pasted it in, then realized—smooth because it said a bunch of things I couldn't argue with, but didn't want to read either.

That's when I knew. I'd hit another blind spot. Had to go fill it myself.

Closed the laptop. Went to open that book I never finished.

Zhou Ye. A middle-aged columnist. Worked as a reporter, spent time in corporate, tried his hand at business. Now makes a living writing. Reads widely, argues hard, writes about things everyone feels but nobody says clearly.