The Barbell and the Forklift

A forklift can lift two tons without breaking a sweat. You cannot. Yet nobody walks into a gym, sees someone straining under a heavy bar, and offers to bring the forklift. Everyone understands the joke instantly. The point of the barbell was never to move the weight. It was to change the person lifting it.

I build AI systems that lift enormous cognitive loads. They read faster than I can, recall more than I can hold, and turn a rough plan into working output in seconds. And I still do my hardest thinking with a pen on paper. Not out of nostalgia, but for the same reason the lifter waves off the forklift. Some work is supposed to change the worker, and handing it to a machine quietly defeats the purpose.

A forklift moves the weight. A barbell changes the person who lifts it. Paper is a barbell. AI is a forklift. The skill is knowing which job is which.

Two machines, two purposes

A forklift exists to move a load from here to there with as little effort as possible. By that standard, modern AI is a magnificent forklift. It carries facts across a million tokens without tiring and hauls a specification into finished code while you watch. When the goal is to move the load, you want the most powerful machine you can get, and you would be foolish to lift by hand.

A barbell exists for the opposite reason. Its whole value is the resistance. The weight is heavy on purpose, because the strain is what builds the muscle, and the muscle is a change inside you that no machine can install from the outside. Understanding works the same way. It is not a file you can copy onto a person. It is strength you earn by doing the lift yourself, which is why writing a hard idea out by hand, slowly, in your own words, leaves you holding something that skimming an AI summary never does.

The mistake almost everyone is making right now is bringing the forklift into the gym. They reach for the most powerful tool for a task whose entire point was the effort, and then wonder why nothing stuck.

What the evidence shows

This is not just a tidy image. The research keeps landing in the same place. In their 2014 study, Pam Mueller and Daniel Oppenheimer found that students who took notes by hand beat laptop users on conceptual questions. The reason was telling. Laptop users typed fast enough to transcribe the lecture word for word, like a forklift hauling boxes untouched. Writers could not keep up, so they had to compress and rephrase, and that strain was the learning. The slower tool built the stronger mind.

The brain studies point the same way. A 2021 experiment using N400 brain responses found that words learned by handwriting showed stronger memory encoding than words learned by typing. A 2024 EEG study by Van der Weel and Van der Meer recorded far wider brain connectivity during handwriting than typing, in the frequency bands tied to memory. I read that one carefully, because it is easy to oversell: the participants wrote familiar words rather than learning new ones, so it shows a richer brain signature, not a guaranteed grade boost. Held to its real claim, it still rhymes with the rest. The hand recruits more of you into the act of forming a thought.

Underneath all of it sits a principle the psychologists Robert and Elizabeth Bjork named the desirable difficulty: the conditions that feel harder while you learn often produce the most durable learning. Difficulty is not a bug in the process. It is the mechanism. Remove it and you remove the result.

Why AI makes the barbell matter more

You might expect better forklifts to retire the barbell. The opposite is true, and it comes down to scarcity. When something becomes cheap and abundant, value moves to whatever stays rare. AI has made capture, recall, and first drafts nearly free. The summary writes itself, the notes organize themselves, the boilerplate appears on demand. What it has not made cheap is understanding, because understanding is that internal change, and no external engine can perform it for you.

So capable AI does not weaken the case for thinking on paper. It sharpens it to a point. The rare and valuable act in a knowledge worker's day is now the deliberate effort of working a hard idea through your own hand, and paper is still the cheapest, quietest place to do that lift. No autocomplete finishing your sentence before you have finished your thought. No model offering the answer that lets you skip the very struggle that would have made you stronger.

Training the loop

A good athlete does not refuse machines. They use each one for what it is built to do, and they never let a machine do the rep that was the whole point of the session. The same discipline maps onto a working day.

• Lift before you load. Do the messy thinking by hand first: the design, the tradeoff, the thing you are trying to reason through. Move to the keyboard only once the idea holds its own shape.
• Let AI carry the load, not build your muscle. Hand the agents the spec, the refactor, the lookup, the draft. These are pallets to move, and they move them well.
• Use AI as a spotter, not a substitute. Ask it to question your handwritten reasoning and probe for gaps, so it sharpens your understanding instead of replacing it.

None of this slows you down where speed matters. It just keeps the forklift out of the one place strength is supposed to come from.

The takeaway

This is not a vote against the machines I help build. It is a claim about which job belongs to which tool. The most automated worker in the room should be the most protective of the one task automation cannot touch, because that task is where their judgment actually comes from. Paper is not where I resist AI. It is where I earn the understanding that makes my use of AI worth anything.

As these systems take over more of what we used to call knowledge work, the people who keep doing their own lifting will pull away from the people who quietly hand it all to the forklift. The pen is a cheap barbell. That is the whole reason it is still on my desk.

What leaders should do

1. Separate thinking time from output time on your calendar, and protect a daily block of analog work where no screen is allowed.
2. Do the design phase of important work by hand before it reaches a tool, so the specification you pass to AI is one you genuinely understand.
3. Point AI at testing your understanding rather than manufacturing it: have it interrogate your notes, not write them.
4. Watch for the team that stops lifting and only hauls. Treat shallow understanding as an operational risk, not a matter of personal style.