The Friction That Makes Ideas Worth Having

What if the most important thing AI gives us is also the thing it quietly takes away?

AI can help us write, think, learn, plan, and connect almost instantly. Yet some of the experiences that make those activities meaningful are slow, awkward, and resistant to automation. They happen in the places where we struggle in public, try ideas that fail, meet people we did not intend to meet, and gradually develop a taste for what is true, useful, or memorable.

This creates a strange paradox. The easier it becomes to produce language, the harder it may become to produce an idea worth saying. The more friction we remove from thinking, the less contact we may have with the reality that makes thinking sharp.

The connection between AI and good copywriting reveals why. The best writing is not merely fluent. It is concrete, falsifiable, distinctive, and shaped by conflict. Those qualities are not decorative features added at the end of a process. They are the residue of contact with difficulty.

The struggle we are most tempted to remove may be the very process that gives our ideas their shape.

The missing room where ideas become real

A third place is somewhere that is neither home nor work. It is a cafe, barbershop, library, neighborhood park, religious community, or corner store. Its value is not just that people gather there. It gives them a setting in which they can become less formal, encounter strangers, and develop a shared local reality.

The same concept can be applied to thinking. We need intellectual third places: environments outside our private minds and official tasks where ideas encounter resistance. A conversation after class, a draft shared with friends, a failed experiment, a long walk with someone who disagrees, or a public attempt to explain something difficult can all perform this function.

AI threatens to become a substitute for these places. It can provide an answer before we have articulated the question. It can simulate a conversation without requiring us to risk embarrassment. It can produce ten possible ideas before we have stayed with one long enough to discover whether it has any life in it.

That convenience is real. It is also dangerous when mistaken for development.

Suppose you are trying to understand a difficult subject. You ask an AI system for an explanation, receive a clean summary, and feel the satisfying sensation of comprehension. But you may have skipped the confusion that would have shown you where your mental model was weak. You have acquired a description without building an internal structure.

Or suppose you want to write an essay. You ask for an outline, then a draft, then a more compelling introduction. The result may be smooth, but smoothness can conceal a problem: you have not yet discovered what you believe. The prose has arrived before the thought.

This is why struggle matters. It is not automatically virtuous. Much struggle is pointless. But the right struggle performs three jobs:

1. It exposes what you do not understand.
2. It forces you to make choices.
3. It gives you evidence about what survives contact with reality.

Without those jobs, AI becomes a machine for producing plausible surfaces.

Why concrete language is a form of thinking

Good copywriting offers a practical test for whether an idea has survived that contact. The test is simple: Can you visualize it? Can nobody else say it? Can you falsify it?

These questions are useful far beyond advertising because they distinguish an idea from a mood.

Consider the phrase regain fitness. It sounds positive, but it does not give the mind anything to hold. What does fitness mean here? Strength? Flexibility? Running? Losing weight? Now narrow it further: getting someone off the couch after six months of inactivity and helping them run five kilometers. The abstract promise has become couch to 5K.

The improvement is not merely stylistic. The narrower version gives us a mechanism, a starting point, a destination, and a way to tell whether the promise has been fulfilled. It can be visualized. It can be tested. It can fail.

The same principle applies to personal and intellectual claims. Someone described as intelligent is difficult to evaluate. Someone who reads on the train every morning is concrete. A person with good values is abstract. A person who returns money when a cashier gives too much change is specific. A product that offers a better way to collaborate is vague. A tool that lets a distributed team send a five minute video instead of scheduling a thirty minute meeting is legible.

Concrete language compresses experience without deleting its evidence.

This matters even more in an age of AI because AI is exceptionally good at generating abstractions that sound finished. It can produce phrases such as innovative solution, seamless experience, meaningful connection, or empowering users with startling fluency. The words are grammatically correct and strategically empty.

The cure is not to ask the machine for more vivid language. It is to ask a prior question: what happened?

Where were you? Who was there? What did they do? What changed? What would a skeptical observer be able to see? If the answer is nothing, the idea has not yet touched the world.

A useful editing exercise is to circle every abstract noun in a draft. Then interrogate each one. What does collaboration look like at 3:00 p.m. on a Tuesday? What does confidence make someone do? What does community sound like when three regulars argue over a football match in a cafe?

The goal is not to eliminate abstraction. We need concepts such as justice, freedom, trust, and beauty. The goal is to connect abstractions to observable particulars. A principle becomes persuasive when it can descend into a scene.

Falsifiability is creative risk

The second test, falsifiability, is even more powerful. A claim that cannot be false cannot teach us much. It may be emotionally pleasing, but it does not put anything at stake.

This explains why strong writing often produces a small physical reaction. We sit up. We wonder whether the statement is true. The writer has placed an actual proposition in front of us rather than wrapping a general sentiment in elegant language.

Compare these two claims:

• AI helps people work better.
• A team that replaces one weekly status meeting with a shared five minute video can reduce coordination time without losing context.

The first is nearly impossible to dispute because it says almost nothing. The second can be measured. It may be wrong. That is precisely why it is more interesting.

Falsifiability creates trust because it signals that the writer is willing to be judged. It also creates memorability. We remember Galileo saying that the Earth moves around the Sun more readily than we remember a statement about humanity’s harmonious relationship with celestial objects. A concrete claim creates conflict with reality, and conflict gives the mind something to organize around.

This suggests a useful model for thinking with AI:

AI can expand the search space, but it cannot decide what deserves risk.

Ask an AI system for fifty headlines and it will give you fifty possibilities. But the important work is choosing one sentence that makes a claim narrow enough to be tested and bold enough to matter. The machine can increase variation. It cannot supply the personal consequences of being wrong.

That personal risk is not a flaw in the process. It is the source of originality. If your sentence could have been written by anyone, it probably reflects no particular encounter with the world. If it could not possibly be disproved, it probably reflects no clear observation.

This is why distinctive writing often begins with an enemy. The enemy need not be a competitor. It can be a different method, a prevailing belief, or a category that has become too small.

A screen recording product can say that it is easy to use. Or it can argue that remote communication fails because meetings are too long and email is too impersonal, then define a new category around asynchronous video messages. The second statement does more than describe a product. It changes the comparison set.

The same move works in personal thinking. Instead of saying that you want to learn more efficiently, you might say that you are trying to stop confusing recognition with understanding. Instead of saying that AI is changing education, you might argue that the central educational scarce resource is no longer information but unassisted attention.

A good enemy creates a boundary. Within that boundary, a point of view can become visible.

The taste problem: what AI cannot struggle into existence

There is a common fantasy that better tools automatically produce better work. History suggests otherwise. Moving from pens to typewriters, typewriters to laptops, or film to digital cameras changed speed and convenience, but not necessarily judgment.

AI is no different. It is a tool, and its output is limited by the taste of the person directing and evaluating it. But taste is not simply a preference for things that look polished. Taste is the ability to notice the difference between what is merely acceptable and what is alive.

That ability develops through exposure, practice, comparison, and rejection. You write a sentence that sounds impressive, then notice that it means nothing. You produce an idea that seems original until someone points out that it is a familiar argument. You test a product message in a crowded environment and discover that it disappears beside seven competing claims.

These failures teach the eye.

A draft viewed alone can feel excellent. Put it in the context where it will actually be encountered, surrounded by notifications, competitors, distractions, and the reader’s fatigue, and its weakness becomes obvious. This is true of an advertisement, a book title, a lesson plan, or an AI generated response.

The practical implication is that evaluation must happen in context, not in isolation. Do not ask only whether a sentence is good. Ask whether it survives the environment in which it must work.

A useful three stage loop is:

Generate, resist, expose.

Generate widely, using AI when it helps you escape the blank page. Resist by selecting, rewriting, narrowing, and defending one idea without outsourcing the judgment. Expose the result to reality through a reader, customer, conversation, experiment, or competing set of messages.

The first stage benefits enormously from automation. The second stage is where taste develops. The third stage is where taste is corrected.

If AI performs all three stages, it can make you faster while leaving you no better. If it performs only the first, it can become a powerful partner in creative work.

A practice for keeping friction productive

The answer is not to reject AI or romanticize difficulty. The answer is to separate productive friction from wasted effort.

Before asking AI to produce a solution, spend a short period making an unassisted attempt. Write the bad paragraph. Draw the rough model. Explain the problem from memory. Predict what will happen. This creates a baseline of your own understanding.

Then use AI as an adversary rather than an oracle. Ask it to identify assumptions, generate counterexamples, find the weakest sentence, or propose conditions under which your claim would be false. This preserves the struggle that matters while removing repetitive labor.

Finally, force the output through four gates:

1. Scene: Can a reader picture a real person doing something?
2. Boundary: What exactly is included, and what is excluded?
3. Risk: What claim could turn out to be wrong?
4. Difference: Why could a specific person, with specific experience, say this better than a generic system?

If a draft fails the scene test, it needs evidence. If it fails the boundary test, it needs narrowing. If it fails the risk test, it needs a sharper claim. If it fails the difference test, it needs more contact with your own observations.

Key Takeaways

• Use AI after your first encounter with the problem, not before it. Make an initial attempt so the tool can strengthen your understanding rather than replace it.
• Translate abstractions into scenes. Replace better communication with a five minute video that prevents a thirty minute meeting. Replace good values with an observable action.
• Make one claim falsifiable. State what would prove you wrong, even if only approximately. Specificity creates credibility.
• Choose an enemy. Name the method, belief, or category your idea rejects. Contrast gives a point of view its shape.
• Test ideas in their real environment. A message that looks good alone may vanish among competing claims, distractions, and ordinary human skepticism.

The deepest danger of AI is not that it will produce bad writing. It is that it will produce acceptable writing so effortlessly that we stop noticing the difference between fluency and thought.

A worthwhile idea usually begins in an uncomfortable place: a failed explanation, an awkward conversation, a stubborn fact, a customer who does not behave as expected, or a blank page that refuses to cooperate. Those moments are not obstacles on the way to the idea. They are the room where the idea is made.

AI can help us leave that room with better tools, more alternatives, and faster feedback. But if it never lets us enter, it gives us language without experience, confidence without evidence, and conclusions without discovery.

The question is therefore not whether AI can think for us. It plainly can perform many operations we associate with thinking. The more important question is whether we are still willing to encounter the friction that gives our thoughts something worth saying.

The future will not belong simply to those who can generate the most words. It will belong to those who can still tell which words have met reality.