Hey! Ankur here, and this is the 21st edition of Lazy AI — 5 mins of reading to help you stay ahead of the AI curve.

Today’s piece is about the AI job panic. The one in every WhatsApp group, every coffee chat, every uncle at every wedding. “Beta, will AI take your job?”

Most people are worried about the wrong thing. And I’ll tell you why.

Last week, Goldman Sachs put out a number. AI is wiping out about 16,000 US jobs every month. Most of them entry-level GenZs.

Then a separate poll came up. 1 in 5 American workers said AI is already doing parts of their job.

Yes, sounds scary. But there are two studies that came out last year, which are not published in the media. Because they’re..well..not dystopian.

Let’s go through them

A research team pulled together 83 different studies that compared AI diagnoses against actual doctors.

The finding: It performed almost as well as physicians overall and non-expert physicians, but it was 15.8% worse than expert physicians

Read that again.

Not “AI is amazing.” Not “AI is useless.” Just, “AI matches the average, but loses to the experts”

You don’t need to be an engineer to talk AI. Subscribe to this newsletter — every other day, you’ll understand AI well enough to be the smartest person in the room at work.

Each issue is just 5 minutes — less than the time you spend doomscrolling before bed. Except, this actually moves your career forward. Join 8,000+ subscribers now!

This one’s crazier.

A research group called METR took 16 senior developers — people with 5+ years on the job, working on their own code. Real bugs. Real features. Nothing fake about the setup.

And they split the work randomly: Half the tasks with AI, and half without. Then they measured the results.

Before starting, the developers guessed AI would make them 24% faster. After finishing, they felt they’d been 20% faster.

The reality? AI made them 19% slower.

And the worst part? They couldn’t even tell. They walked out of the study sure AI had helped them. But the numbers told a different story.

Well, here’s where it gets interesting.

Both studies were in different fields, but showed the same pattern: AI does great work at average. But it gets worse the more expert you are.

Because AI has pattern. Not taste.

You see, Pattern is what AI does. It predicts the most likely answer based on everything it has read. It “predicts”. The “most likely” answer.

And that’s the problem.

Because taste is what you do. You know the answer is wrong even when you can’t explain why. Built from being wrong enough times — in exactly the right ways.

Think about the finance guy on your team. Not the junior, the senior one — the one who’s seen 10 budget cycles. He looks at a number in a model and says, “this can’t be right.” He can’t always tell you why in that moment. He just knows the number is off. Two hours later, he finds the broken formula three sheets deep.

That’s taste. AI doesn’t have it.

Or your best marketer. She reads the AI-written brief and changes three things. Not because the brief is wrong on paper. Because she knows that for this audience, a short subject line will work better. She knows this product sells better with a story than a stat. She knows the room. She can’t always explain it. She just…knows.

That’s taste too.

And AI can’t see any of it.

I write this newsletter every week. I use AI for it. A lot.

But you know what AI keeps getting wrong for me? The numbers. It will put in a stat that doesn’t exist, or a broken link, or a study that just doesn’t have the number we’re basing our research on. The writing looks fine and the flow is clean. But the numbers are wrong.

So I check every single number and every single fact before I finalise the draft.

The other thing it messes up is subject lines. Claude writes subject lines that sound like every other AI newsletter on the internet. Sometimes cringe, other times forgettable.

So that’s the part I just do myself. Because I know what you guys will end up opening.

Claude doesn’t have that taste yet.

Look, I’m not saying that AI is sloppy. I use it every day. In fact I even burnt through Claude’s tokens last month, in spite of being on the paid plan 🥲

But the part of my work that actually matters — knowing which number is wrong, which subject line you’ll actually click on, which sentence to cut — none of that comes from AI.

It comes from writing this 21 times in a row. And being wrong enough times to know the difference.

You don’t need to be an engineer to talk AI. Subscribe to this newsletter — every other day, you’ll understand AI well enough to be the smartest person in the room at work.

Each issue is just 5 minutes — less than the time you spend doomscrolling before bed. Except, this actually moves your career forward. Join 8,000+ subscribers now!

The panic is not wrong. Average work IS in trouble. Entry-level jobs ARE getting cut.

But every tech wave before this — calculator, excel, internet, mobile — went the same way. The people who survived were the ones who spotted where it fell short and got really good at that.

So the question is not really “will AI replace me?”

It’s “have I built enough taste to know when AI is wrong?”

Because the day AI develops taste, it’s not AI anymore. It’s AGI. And that’s when the problem really begins.

But that’s a story for another day.

Until then, this gap between pattern and taste is what you need to identify. Because that’s exactly where you can excel.

If this got you thinking, share it with someone who’s quietly worried about their job and doesn’t know what to do about it.

See you next time..

Cheers,

Ankur