Filip Hráček / text /

A.I. is a printed birthday card train to Paris

A new technology arrives and everybody’s like, wow, this will save so much time! We’re going to work a lot less now, aren’t we.

For example, when they linked cities with railway in the 19th century, a salesman from Prague who regularly visited Paris for work could have said:

... What would have taken me a week is now an overnight journey! I’m still going to take a week to go to Paris but I’ll have a little vacation along the way!"

On the surface, this might be a completely logical thing to say. You do the same amount of work in a fraction of the time. Why not take some of that saved time and live a little?

The problem, of course, is context. The salesman is not the only person in the world who knows about trains. Everyone knows. His boss. His peers. His clients. His competition.

So, instead of the new technology becoming an unprecedented advantage, it becomes table stakes. Pretty soon, everyone goes by train, and those who insist on taking the horse-drawn carriage are out of work (or change their mind).

Put a pin in it, I’ll get back to this point later.

---

A young lady celebrates her birthday and, apart from a gift or two, she also receives birthday cards from her parents.

The dad writes the card by hand, and adds a little drawing. The mom writes it in MS Word, adds an emoji, and prints it out on the home printer.

The text is identical.

Taken with the shallowest reading, the two cards are equivalent. Mom’s card even has some advantages: it’s cleaner, more readable, and the drawing is of a higher quality (made by a professional illustrator). She spent less time making it. Win-win, right?

But you’d probably understand if you found out that the young lady cherishes dad’s card more than mom’s.

The reason is, once again, in context. The textual content of the two cards is the same, but the message is different.

---

As a university student, I was once a co-organizer for a tiny music festival we held on our campus. I remember talking to a friend who was also part of the organizing team while a famous local folk singer performed on stage. The artist was, presumably, sitting on a chair with the guitar on his lap, strumming and singing into the microphone.

I say “presumably” because neither of us could actually see the singer: the place was packed. We could only hear the music.

My friend smirked and said:

Well, I’m happy. In fact, I’d be happy even if it was an mp3 player sitting on that chair, playing these same songs.

We both laughed.

Because that is absurd. Of course you want the actual singer there, and not some MP3 player. Even if you can’t see him. Even if that MP3 player produces better sound (the songs played from the player were produced in a studio, after many takes, using the best recording equipment, and carefully mixed for days by an audio engineer). No, you pay the price of the ticket and you stand in the mud because you want to hear the man himself.

---

Imagine teaching a feral child to integrate. The kid has never learned how to count, or at least not in a way a civilized kid learns that in school. The child is feral. It was raised by wolves — and wolves, one must presume, aren’t good math teachers.

Nevertheless, the kid is of human intelligence. It can recognize patterns and apply methods. It can learn the power rule, the reciprocal rule, the exponential rule and the trigonometry rules. It just needs to memorize some names and symbols, learn basic math (e.g., to apply the power rule, you just need to be able to add 1 to a number) and get some practice in combining them.

In a month, you have a kid that can ace a test meant for university students. It can take scary-looking mathematical formulas and correctly convert them to other mathematical formulas. All that without ever learning things like trigonometry, exponentiation, or even multiplication. Heck, to be able to add 1 to a number, the kid probably doesn’t even need to really understand addition.

Has the feral child leap-frogged all these other kids who have been studying math several hours a week for years? Of course not. It’s a party trick. The kid’s understanding of integration is as shallow as it can be. It can’t actually apply the knowledge outside a test.

This is because skills are layered. To truly intuit a higher-order skill, you have to be solid in the skills it stands on. Integration is a powerful tool, but it’s useless in the hands of someone who can’t multiply.

Yes, there are shortcuts. You don’t need to be able to derive the power rule of integration. Most students can’t do it. But they at least understand the basics of functions, equations, arithmetic. Their understanding of integration stands on something.

Similarly, if you want to be able to author great articles, you can’t just be a good “article planner”. You need to be able to string words into sentences and sentences into paragraphs. You need to understand a thousand little tricks that writers do, unconsciously, with language. Yes, there are shortcuts. You don’t also need to be a good calligrapher, nor do you need to understand the mechanics of keyboards. But there are skill layers you can’t skip.

Also, similarly, if you want to write good software, you can’t just operate on the level of a Team Lead without first learning to program. Yes, shortcuts abound. You don’t need to be able to convert from binary to hexadecimal by hand, nor do you need to learn assembly for writing web apps. But you can’t expect to be a good software architect (in the broadest possible sense of the word) without having deep experience with writing software, line by line, function by function.

---

These are the two interrelated points that I wanted to make today.

Number one. A.I. is a productivity tool. That means it will not make our workdays shorter. It will only make them faster. That’s not nothing, of course, but don’t trust anyone who tries to tell you that A.I. will give us more leisure time. That’s not how new tech works. (If it was, we’d be working 1 day a week by this point.)

Number two. A.I. is a productivity tool. Don’t fall into the trap of trying to replace yourself with A.I. Use it as a tool, not a replacement.

It’s already possible, even today, to replace much of your role as a productive member of society with A.I. For example, I know people who are experts in their fields, and who have recently succumbed to the comfort of A.I. writing. Where they once wrote a few authentic articles per year (which made them the experts in the eyes of their respective industries), you can now read several of their articles per week. These articles are long, well thought out, without grammatical mistakes. Sometimes, you can’t even tell they were written by a tool — the A.I. uses some of the original person’s style.

But while this must feel wonderful to the experts today, I’m convinced they’re unwittingly shooting themselves in the foot. They’re replacing themselves with an MP3 player and, sooner or later, people will stop listening. They’re forgetting how to multiply, and sooner or later, their prowess will evaporate. They’re giving people printed cards, and sooner or later, people will feel the difference.

This is why, if you read something on the internet from me, Filip, I can guarantee I didn't just instruct an A.I. to write it. Even if it means there are errors and the whole thing is less polished and it takes more time to write.

I’m not doing this because I love doing more work, nor because I’m a contrarian Luddite. I’m merely trying to save my long-term livelihood from my short-term laziness.

Aside: I realize how hilarious it would be if I revealed, at the end of this article, that it was all written by A.I. But although I love a good plot twist, no such thing is coming.

---

So, on one hand, refusing to use A.I. at all is a bad idea. People aren’t being lazy for opting to use A.I. — they’re simply adapting to the new table stakes. If you've been avoiding LLMs like the plague, it’s probably time to revise your approach. You might not be able to take the horse-drawn carriage to Paris much longer.

On the other hand, going all in — and basing your skill set fully on A.I. — is also a bad idea. By outsourcing your craft to an LLM, your grasp of your expertise is slipping, too. You’re quickly becoming a feral child that can integrate.

And as A.I. is ceasing to be a new, unknown technology, the party trick is getting old.

— Filip Hráček
November 2025