Listen to this post: Apple Podcasts | Spotify | YouTube
A five-part series on working with the robots. Part 1, Part 2
LLMs are convergent thinkers. They take a problem and find the most probable answer. Average the inputs, smooth the edges, hand back something that is likely to satisfy the user without ruffling any feathers. It’s running statistical probabilities inside the existing data.
Humans, at our best, do the opposite. We diverge. We break the frame.
There’s a study from the University of Sydney where researchers gave participants the famous nine-dot puzzle.
Before you keep reading, try it yourself.
Connect all nine dots using four straight lines without lifting your pen.
Need a hint?
Think outside the box. Literally.
But what box?
Most people invent one. They see the nine dots and unconsciously draw a square around them. They invent a rule where none exists.
SPOILER ALERT: SOLVED PUZZLE BELOW
In the control group, nobody solved it. Then researchers used transcranial direct-current stimulation to temporarily suppress the brain’s “top-down” filters—the rules-following part. Fictional rules removed, over 40% cracked it.
The puzzle didn’t change. The rules didn’t change. People just stopped obeying a boundary that never existed.
I gave the puzzle to my kids. My son eventually solved it with that hint: “Think outside the box.”
But the thing I noticed wasn’t whether the kids got the answer.
It was how long they wanted to sit with it. My daughter wanted the puzzle and the hint in quick succession. But then, she also wanted the answer.
My son wanted the hint as well, and then disappeared. He noodled. Came back with a few failed versions. Broke the rules once or twice. Went back to the drawing board. And eventually, he found the answer.
This was no shock to me. My daughter is a type A with straight A’s. My son is divergent as all hell.
The frame breaks, the boundary disappears, when you sit with the problem long enough. But the modern world trains against sitting with problems. Every unanswered question feels intolerable.
But robots can’t sit either. They answer quickly. Too quickly to wander.
Humans have an advantage there. We wander. Stare out the window. Circle a problem for three days. Get stuck long enough for something unexpected to happen.
LLMs can’t break frames. No neurobiology. No boredom. No wandering. No long quiet incubation where unrelated ideas suddenly collide.
They are built to stay inside the lines. As they automate convergent work—and they will, more and more—the human edge becomes a different kind of work. The lateral spark. The connection nobody saw.
The term “soft skill” is a 1960s era Army term. The military separated training into two buckets—“hard skills,” meaning anything that involved operating a piece of metal: tanks, guns, radios, vehicles. And “soft skills,” meaning everything else. Leadership. Communication. Reading a room. Building trust.
The label stuck. But notice what happened. The skills that mattered most for succeeding in any role—the social, emotional, behavioral ones—got branded as the lesser category. Because they didn’t involve metal.
The “+” is a soft skill. And like every soft skill, you’ve gotta use it or you’ll lose it.
The Train Kid
There’s a viral clip online of a teen on a subway platform “helping” push the train.
He plants his hands on the side of the car, leans against it with all his weight, and brings the train to a stop for waiting riders. He holds the doors open, waves them aboard, and makes sure they close securely before giving a sturdy push to the train, getting it moving towards its next destination. He brushes off his hands and walks away from the tracks, a job well done, a mission accomplished.
Of course, he wasn’t needed for any of it.
But the kid feels the sensation of contribution. The high of usefulness and participation.
Sometimes, working with the robots feels just like that.
The problem isn’t just that they make you lazy.
The problem is that they make you feel productive while being lazy. They make you feel like a centaur while you’re actually the kid pushing the train. The work gets done—and it gets done well—but you didn’t really do it.
The robot did the work.
Like autotune drift in music.
Like Instagram face.
Different people, different surgeons, different cities—and somehow everybody converges toward the same cheekbones.
At first the tools smooth out the edges. They tighten things up. Make everything a little cleaner, a little sharper, a little more optimized.
Then one day everybody sounds the same. And nobody is quite sure when it happened.
I’ve heard editors talk about this. Their concern is a valid one. The work is all the same. They’ll take on projects with seven different writers and the writing all sounds the same. It’s all been passed through LLM editors, it’s all been offered the same structural notes, the same suggested phrasing. The writers swear it isn’t true, that the writing is theirs and theirs alone.
Like the frog in the pot, the temperature changed so slowly they never noticed they were cooked.
The writers can’t hear it anymore. They can’t see how their writing coach has changed them over time. Has changed thousands of other writers as well.
That’s the danger. Just like the kid at the subway station, they can’t tell—or don’t want to believe—that they aren’t actually controlling the train. Feeling productive is not the same as being productive. Feeling original is not the same as being original.
The trap is making it look like work without doing the work. That’s true for the kid on the platform and it might be true for your career as well.
So how do you know which one you are?
Ask yourself these three questions, and try to give an honest answer…without your robot.
If the AI vanished tomorrow, could you still do the work? Maybe not at the same speed or volume—but could you do it? If yes, you’re collaborating. If no, you’re the kid on the platform.
Whose idea was it? Who wrote the first sentence? You, before the AI saw it? Or the AI, before you knew what you wanted? The first version, that early draft, it needs to be your contribution. Otherwise, every draft that follows will carry the robot’s watermark.
Hold your work up next to someone else using the same tools. Can you tell them apart? If yes, your voice survived. Congrats! If no, the tools are doing your thinking, and your work is blending into the background.
The robots will keep getting better at converging.
Your job is to keep diverging.
If you enjoyed this post, please share it with a friend.
What do you think?