In April of 2025, I wrote that people with better taste would always get more out of the same AI tools that everyone else is using. But what about people with more education?

By now, I’m guessing you see people using LLMs in the workplace every single day of your life, am I right? From emails, to pitch decks, to your CTO’s latest version of Flappy Bird she made with your company logo, who’s getting the most utility out of these tools, really?

Since I wrote my article, some scientific studies have been done on exactly this topic.

Here’s one such experiment, quoted from the original paper on NBER.org:

“…participants receive an email from their boss, and to answer it they must analyze several sources of information. They must write a reply to the email that diagnoses the problem and proposes a solution.”

Sounds pretty familiar, right?

Participants were either given ChatGPT to write their response, or asked to do it unassisted.

“Without AI, higher-education participants outperform lower-education participants by 0.548 standard deviations; with AI, the gap falls to 0.139, closing about three-quarters of the initial difference.”

That means that less educated participants performed nearly as well, thanks to AI! That’s massive!

What does this mean?

It means that people with less education and who are worse at writing get more initial benefit out of AI’s capabilities.

But not so fast:

In another experiment from Kenya:

640 Kenyan entrepreneurs were given an AI business mentor and tracked on real revenue and profit for months.

“Subsample analyses show that low performers did nearly 10% worse because of the AI assistant, whereas high performers may have benefited by over 15%.”

So high-performers still get more out of these tools on balance.

Now that’s a lot of facts and numbers, but what does it all mean?

On its face, you might say that GPT’s help is a massive win for people who lack formal education—for people who aren’t strong writers or problem solvers to begin with. But there’s more nuance here, that my own experience bears out.

While an AI tool will help a low performer write a better email in a one-off test, this result can be deceptive. For that moment or that email, an employee appears to perform much better than their own intellectual capacity. But if you were to call that person on the phone, they wouldn’t know or remember the beautiful email they just sent.

I’ve witnessed this extensively in my own career, now that I receive dozens (hundreds?) of AI-generated emails in serious business situations every single day.

These long-winded AI slop emails still don’t tell me what anyone actually thinks, feels, or truly believes—I just know what AI wrote for them.

So this means as business owners, that we can’t trust a well-crafted email. The best way to know what someone is truly capable of is to call them on the phone. You might be surprised at how completely different it is to simply talk to a person unaided by AI, and how refreshing and honest it feels in comparison to reading their latest GPT soup.

…is what my two year old son said to me when he handed me a regular cup full of apple juice.

For someone who’s used to sippy cups, a regular cup is “broken” because it doesn’t work the same way. He couldn’t get what he wanted out of it.

But you and I know, it’s just a cup.

How many things in life are like that?

“The golf club isn’t working!” But Scottie Scheffler just used it to make an eagle.

“The baseball bat isn’t working!” But Shohei just knocked one out of the park with it.

“AI is a waste of time and has no practical purpose!”

…

Many times in life, it’s the operator that needs an upgrade, not the tool.

I want you to wrap your head around something: Mark Zuckerberg built the first version of TheFacebook, pre-AI, by himself.

This means: With AI, a reasonable power user can replicate the functionality of early Facebook in a weekend of coding.

Why does this matter? There’s a reason we all fell in love with Facebook in college: It was a way to keep up with friends, and see when their birthday was, what color their beer bong was, and so on. That value never left.

But over time, a once useful app got surrounded by so much bloat, ads, propaganda, and other garbage that the original intent is almost completely lost.

Today, the only thing I use Facebook for is birthday notifications.

But why do we need all the other crap around this basic-yet-still-important functionality? Why do we have to accept the evolution of Facebook? After all, the only reason they want you to login is to bombard you with ads.

So I built my own software app to remind me to keep in touch with all my most important friends, including birthdays, special days, interests, call-logging, and more. It reminds me to keep in touch with my most important people, something I’m admittedly terrible at without help.

In a week of coding with AI, I made something more valuable to me than Facebook was for the last several years.

Now, I won’t let friendships die anymore, unless I want them to. Cough.

I get a daily email digest. My interface is a lot cleaner, and nary an ad in sight!

A reminder that we don't need to accept ANY existing app that no longer serves its purpose.

One of the common criticisms I hear about AI for coding is that the tools are good for hobbyists, but they aren’t up to the task of real commercial work in high-stakes situations.

While I agree with the sentiment behind those statements, if we look at the real world today, we see something very different:

Boris Cherny, responsible for actually creating Claude Code within Anthropic, said in a recent video:

“It’s writing almost all [of the internal code at Anthropic]… for me personally it’s been writing 100% of my code for at least 6 months. The work of engineering has just completely changed. I feel like I suddenly have superpowers… engineering has never been this fun.”

At Spotify, one of the world’s most beloved apps, their own internal report says that “Today, more than 99% of our engineers use AI coding tools every week, 94% report that AI has made them more productive, and we’re seeing a 76% increase in pull request frequency, with the vast majority of PRs authored by a developer working alongside an AI agent.”

And on a personal level, in addition to being amazed daily at what I can code with these tools, having made dozens of personal apps and two major apps that power my entire business and team of 10+, I also see that the engineers whom I respect the most in this world, including one who literally wrote dozens of books on how to code over the last 30 years, now don’t write so much as a line of code by hand anymore. And yet their genius isn’t diminished but enhanced.

It’s clear to me that these tools are not just novelties and gimmicks used to create AI slop (although there is plenty of that going around), but in the right hands, they are literal magic for seasoned developers everywhere.

If you’re a business owner wondering how/where to use these tools in your own operations, my suggestion is to stop doomscrolling and listening to what these tools can’t do, and start looking at the extremely talented people who know how to code legitimately, and study what they’re doing.

…what you think of as your greatest “superpower” will shortly become your biggest weakness.

It’s not just your secret weapon, it’s everyone else’s, too.