The blank page has never been efficient.
That is part of its value.
It confronts us with the distance between having a vague impression and making a claim. It reveals whether an idea has structure, whether the evidence supports it, and whether we believe what we are trying to say. The discomfort is not incidental to writing. Often, it is the mechanism through which thought becomes clearer.
Generative AI can remove that discomfort in seconds.
This is usually described as a productivity gain. Sometimes it is. But if the machine arrives too early, it can also interrupt the process through which a person develops judgment, conviction, and a voice of their own.
The question is not whether people should use AI.
The question is when AI should enter the act of thinking.
The Sequence Matters
In a recent New York Times essay, design educator Debbie Millman describes students using AI to synthesize and augment thesis writing they were expected to produce themselves. Her deeper concern is not the obvious one about cheating. It is that students can now bypass the difficulty through which they learn to form and defend an argument.
That distinction matters.
If AI helps a person examine an idea they have already begun to form, it can expand the work. It can challenge assumptions, expose missing evidence, suggest alternatives, and make revision more rigorous.
If AI supplies the first interpretation, the first structure, and the first language, it does something very different. It shapes the field before the person has had a chance to discover what they notice, doubt, or believe.
The same tool can strengthen judgment or prevent its formation.
The difference is often sequence.
What the Blank Page Actually Does
Writing is frequently treated as the final packaging of thought. First we know what we mean; then we put it into words.
Real writing rarely works that way.
We discover the weakness of an idea when we cannot explain it. We notice a contradiction when two paragraphs refuse to connect. We find the actual question halfway through a failed answer. We revise a sentence and realize that the revision changes our position.
The page is not merely where thought is recorded.
It is where thought is tested.
This is why a first draft can feel slow, embarrassing, and strangely personal. The writer is not only producing text. The writer is encountering the limits of their own understanding.
An AI-generated draft can conceal those limits before we have learned anything from them. Its fluency gives an unfinished idea the appearance of completion. Its structure can make a borrowed position feel like a considered one. Its confidence can arrive before ours has been earned.
The result may be better prose and weaker authorship.
Not All Difficulty Is Valuable
There is a risk of becoming sentimental about struggle.
Difficulty is not automatically educational. Some of it is simply friction. Correcting spelling, reformatting citations, transcribing an interview, reorganizing notes, or turning a repeated explanation into a cleaner paragraph may consume time without developing much judgment.
AI is excellent at removing this kind of friction.
The important distinction is between mechanical difficulty and formative difficulty.
Mechanical difficulty slows the expression of thought. Formative difficulty changes the person doing the thinking.
Formative difficulty includes:
- deciding what the real question is;
- staying with ambiguity before reaching for an answer;
- choosing which evidence matters;
- discovering the assumptions beneath a position;
- finding language precise enough to expose disagreement;
- and accepting responsibility for a conclusion.
These are not inefficiencies to eliminate. They are capabilities to develop.
The danger is not that AI makes work easier. It is that AI makes the two kinds of difficulty harder to distinguish. A person asks for help removing friction and unknowingly removes the part that was forming their judgment.
Readiness Is a Capability, Not an Age
We do not need an age limit for AI.
Age is an unreliable proxy for intellectual readiness. A young person can have a disciplined relationship with language, while an experienced professional can surrender judgment to the first polished answer a model produces.
Readiness is better understood as a set of capabilities.
Can you form a position before asking the machine for one?
Can you detect when fluent language is hiding weak reasoning?
Can you reconstruct the argument without the generated prose in front of you?
Can you explain why a sentence belongs, what evidence supports it, and what would change your mind?
Can you reject an attractive answer without needing another model to give you permission?
Most importantly, do you know the sound of your own voice well enough to notice when it has disappeared?
A person who can do these things can use AI as an interlocutor rather than an oracle. They can argue with it, redirect it, and refuse its false polish. The machine becomes part of a thinking process the person still owns.
Without those capabilities, augmentation can become substitution while still feeling collaborative.
Protected Thought
The practical response is not prohibition. It is to preserve periods of thought before assistance.
We need protected spaces in which a person encounters the problem before the model does.
In education, that may mean beginning an assignment in class, discussing a reading before generating a summary, or requiring students to submit the evolution of an argument rather than only its polished endpoint.
At work, it may mean asking participants to write their own position before an AI-assisted strategy session. It may mean separating the human framing of a problem from the machine's exploration of solutions. It may mean reviewing the assumptions in an AI-generated plan before debating its recommendations.
The principle is simple:
Do not let AI occupy the first moment of contact with every difficult question.
That first moment is where attention reveals itself. What did you notice? What felt uncertain? Which possibility occurred to you before the available answers narrowed the space?
If every encounter begins with synthesis, we may become excellent at evaluating options we would never have generated and increasingly unable to originate the frame in which those options appear.
A Discipline for AI-Assisted Work
A useful practice can be built around four stages.
1. Encounter the problem alone
Read, observe, and take notes before prompting. Allow the problem to remain unresolved long enough for your own questions to emerge.
2. Form a provisional claim
Write what you currently think, even if it is incomplete or badly expressed. The goal is not polish. It is to establish an intellectual position from which you can engage.
3. Invite the machine to challenge, not replace
Ask for counterarguments, missing evidence, alternative structures, or weaknesses in the reasoning. Use AI to create productive resistance around a thought that already has a human origin.
4. Pass the ownership test
Remove the generated text and explain the argument in your own words. Defend the claims. Identify the evidence. State where uncertainty remains.
If you cannot do that, the work may be in your document without yet being yours.
This sequence will not fit every task. Sometimes the work is routine and formation is not the objective. But for learning, strategy, leadership, research, and consequential decisions, preserving the sequence matters.
The Right We Should Preserve
The blank page is not sacred because writing must be painful.
It matters because people need places where their first response has not already been anticipated, completed, and improved by a system trained on everyone else's language.
AI can help us write better. It can help us test ideas, cross disciplines, escape habitual phrasing, and see possibilities we would otherwise miss. Those are real gains.
But the ability to use AI well depends on capacities the machine can also quietly prevent us from developing.
We need the ability to remain confused without treating confusion as failure. We need the patience to form a thought before optimizing its expression. We need enough experience with our own uncertain language to recognize the difference between fluency and truth.
The right to a blank page is the right to encounter our own minds before a machine offers to complete them.
We should protect it.
Form before you augment.