The AI Era Rewired How I Ask Questions — How Prompts Reshaped the Topography of Thought

After I started using Claude Code, my sentences got longer. The time I spent figuring out what I actually wanted, before asking for code, kept growing. And then one day I picked up a book of poetry at the bookstore. AI had become a mirror reflecting the shape of my own thinking.

1. My Sentences Got Longer

I noticed the change in a small moment.

I was writing a prompt to give Claude Code a task, and it wouldn’t fit into a single sentence. I started typing “refactor this file” and my fingers stopped. What direction should the refactor take, what are the constraints, what should the final artifact look like? A handful of failed attempts had taught me that unless I sorted these out first, whatever came back would be useless.

Ask AI vaguely, and you get a vague answer. Say “fix it however you see fit,” and the AI really does fix it however it sees fit. Standing in front of the prompt window, I realized for the first time how reckless it was to expect an answer from AI when I didn’t even know what I wanted myself.

So the sentences got longer. The time spent ordering the question outgrew the time spent writing the code. At first I thought this was inefficient. Before, I’d just start hammering the keyboard; now I sit blankly for a while, thinking about what I’m even trying to build. And strangely, this “inefficiency” raised the quality of the output. The more precisely I wrote the prompt, the more accurate the AI’s output became — and more importantly, what I wanted to build became clearer inside me.

I wondered if this experience was mine alone. Were other developers, other AI users, going through the same shift?

2. Prompt Literacy — The Question Is the Thought

Andrej Karpathy dropped a line in 2023:

“The hottest new programming language is English.”

It sounds like a joke, but it’s accurate. In the AI era, the most important programming skill is conveying intent precisely in natural language. Naming variables well, splitting functions cleanly — those traditional coding virtues have now extended into the domain of prompt writing.

This isn’t a foreign concept. Any programmer knows Rubber Duck Debugging: the technique where explaining a bug to a rubber duck makes you find the answer yourself. The heart of rubber duck debugging is that the act of explaining itself activates metacognition — thinking about your own thinking.

Writing a prompt to AI is a buffed version of rubber duck debugging. The rubber duck doesn’t talk back, but the AI does — sometimes sharply. When the feedback comes back as “your requirements contradict each other” or “this part isn’t clear,” you reread the prompt and discover holes in your own reasoning.

A 2024 paper at ACM CHI captured this phenomenon academically. The researchers argued that interaction with generative AI demands “meta-awareness and ambiguity tolerance” from users. Knowing that AI output isn’t always accurate, users must review it critically and reflect on whether their question was sufficiently clear. The process itself becomes a cognitive workout.

Mitchell Hashimoto (founder of HashiCorp) has said that using AI coding tools made the ability to decompose work the central skill. Breaking a big problem into smaller units, defining the expected outcome of each, designing the order — this is programming in the AI era. Thinking about architecture has become more important than typing code.

How you ask is how you think. The prompt isn’t a command sent to AI; it’s an instrument for structuring your own thought.

3. AI Is a Mirror — The Cognitive Mirror

Stanford philosopher Shannon Vallor wrote, in her 2024 book The AI Mirror:

“AI is a mirror that can show us what we are now, but it cannot tell us who we might become.”

The metaphor maps exactly onto my experience. Write a vague prompt and you get a vague result. Write a logically tight prompt and you get a refined result. The quality of AI’s output mirrors the quality of my thinking, line for line. The way someone standing in front of a mirror shows up in the glass.

Psychology Today analyzed this and reported that conversation with AI has the effect of generating an “internal cognitive map.” In the process of explaining your thinking to a chatbot, the structures and biases of your reasoning — ones you hadn’t even noticed yourself — come into view.

A 2025 paper in Frontiers in Artificial Intelligence formalized this into the “Cognitive Mirror” framework. According to the framework, AI systems both reflect and amplify human cognitive patterns. Clear thinking comes back clearer; confused thinking comes back more confused.

The crucial thing is to know the mirror’s limits. As Vallor warns, the AI mirror reflects only the current you. Future possibility, ideas not yet formed, the territory of intuition — these don’t show up in the mirror. What AI displays is the range of what you can already express in language. Beyond that is still human work.

So the effort to write good prompts isn’t merely skill acquisition. It’s the effort to raise the quality of your own thinking, an attempt to sharpen the figure in the mirror.

4. A Developer Started Reading Poetry

The most unexpected change in me was this: I started reading poetry.

An engineer who’d read technical documentation and code for twenty years, one day in a bookstore, picked up a collection by Yun Dong-ju. I bought an introduction to philosophy. I started spending time on foreign languages. At first I didn’t know why I was doing this. Burnout? A midlife crisis?

Only later did it click. Talking with AI had made me sensitive to the precision of language. Even with the same meaning, the AI’s response shifts depending on which word you choose. “Quickly” and “efficiently” are different; “simply” and “just the essentials” are different. The nuance of a single word changes the result. The experience awakened a sensitivity to language itself.

And evidence is mounting that this change isn’t mine alone.

AI training-data companies Scale AI and Appen are hiring poets and creative writers at $50 per hour to evaluate and improve the linguistic quality of AI models. The fact that the key labor force of the AI era is poets rather than software engineers is telling. To make machines speak better, humans first have to know how to read well.

Cornell professor Laurent Dubreuil has argued that AI is resetting the qualitative baseline of the humanities. In an era when AI cranks out average essays, poems, and analyses, human writing and thinking acquire value only by reaching depths AI cannot touch. Paradoxically, AI’s arrival is raising the value of humanistic capability.

These conversations are also growing in developer communities. “In an era when AI writes the code, where does a developer’s identity lie?” “The transition from a coder typing code, to an architect designing systems, to a thinker who defines the problem.” Discussions about craftsmanship and the meaning of creation are running hot.

A developer who reads poetry. An engineer who studies philosophy. In the AI era this isn’t a hobby — it may be the extension of a core competency.

5. Counterargument — Is AI Degrading Thought?

Here an uncomfortable question can’t be avoided. Does AI really deepen thinking? Or is it doing the opposite — letting thought atrophy?

In November 2025, the Harvard Gazette ran an article warning of “cognitive atrophy.” The more we depend on AI, the less we think for ourselves. The same pattern by which mental arithmetic eroded after the calculator’s arrival, the piece warned, is now unfolding in critical thinking and creativity.

Anthropic’s own research, published in January 2026, was more concrete. Developers using AI coding tools showed a 17% drop in code comprehension. They were using AI-generated code without sufficiently understanding it. That an AI company would publish such a warning about its own product carries weight.

MIT Media Lab’s brain-monitoring research deserves attention too. A substantial share of participants who wrote with AI assistance could not accurately remember an hour later what they had written. The boundary between AI-authored and self-authored sections blurred, and memory of text that wasn’t theirs faded fast.

Public opinion reflects the same concern. A Pew Research Center poll from June 2025 found that 53% of Americans believe AI will worsen human creative thinking.

This data can’t be dismissed. It collides head-on with my own sense that AI deepens thought. How should we read this?

The hinge is how you use it.

It’s the difference between delegation and engagement. Delegate the thinking to AI — “handle it however,” “write something passable” — and the cognitive muscle weakens. But engage AI as a thinking partner — organize your own thought, critically review its output, exchange feedback and iterate — and the cognitive muscle gets stronger.

The same tool can produce opposite effects depending on the person and the usage. This isn’t unique to AI. The internet, social media, smartphones — the same pattern held. Tools are neutral; the user’s intent decides the outcome.

6. In Search of the Pattern — People Living This Experience

A pattern is beginning to emerge among people who deliberately use AI to go deeper, not shallower.

There are roughly two camps: people who write first and refine with AI, and people who hand the task straight to AI. The former organize their thinking in language and then deploy AI as editor or critic. The latter let AI generate the output from scratch. The cognitive effects are dramatically different. The first keeps the steering wheel; the second hands it over.

Simon Willison (core Django developer, LLM tooling specialist) has set a clear rule. “Never let AI speak in my voice.” Use AI for research and drafting, but every sentence that finally reaches the reader is written by his own hand. The rule is a declaration: keep AI as a tool, and keep the identity of his thinking and expression his own.

Darius Foroux (writer, productivity specialist) is more blunt. “I want to out-human AI writing.” His goal is to write at a level of experience, emotion, and insight AI can’t reach. AI’s arrival, he said, raised the bar for his own writing.

Wharton professor Ethan Mollick offers a different angle. “Just do stuff with AI for about 10 hours.” Save the judgment for after. Before theoretical analysis or ethical debate, you have to feel it firsthand to find your own way of using it. Ten hours of experimentation beats a hundred hours of discussion.

The three approaches differ, but they share something. They all use AI as an extension of their thinking, not a replacement for it. The wheel stays with the human.

Statistically, this isn’t a fringe phenomenon either. As of 2025, 52–56% of American adults use AI, and the figure reaches 76% among those under 30. The effect of AI interaction on thought is no longer an early-adopter story; it’s a population-scale cultural shift.

7. Language and Thought — A Sapir–Whorf Hypothesis for AI

Let’s push one step further. If conversation with AI changes thought, doesn’t that connect to the old hypothesis that language shapes thought?

The Sapir–Whorf Hypothesis holds that the language we use shapes the way we think. The strong version (language determines thought) is largely rejected by academia, but the weak version (language influences thought) has held up across studies. For example, English speakers, who represent time horizontally, and Chinese speakers, who represent it vertically, do differ in their spatial cognition of time.

The interesting claim is that this hypothesis applies to programming languages too. APL’s creator Kenneth Iverson titled his 1979 Turing Award lecture “Notation as a Tool of Thought,” arguing that notation is an instrument of thinking. The structure of a programming language shapes the thought patterns of the programmer who uses it. Lisp programmers think recursively, SQL programmers think in sets, Haskell programmers think in type systems.

So what about conversation with AI? Isn’t writing prompts analogous to learning a new “language”?

When we speak to AI, we structure thought in a peculiar way. We provide context first, declare constraints, define the expected outcome. This is unlike everyday conversation, unlike code writing, unlike essay writing — a new kind of linguistic act. If this new linguistic act is reshaping our thought patterns, then a modern variant of the Sapir–Whorf hypothesis is playing out.

Oxford philosopher Luciano Floridi, in analyzing the epistemic shift of the AI era, has offered two concepts. One is “semantic pareidolia” — meaning-pareidolia. Like seeing a face in clouds, humans tend to read into AI output understandings and intentions that aren’t really there. The other is “onlife” — a mode of life where the line between online and offline has vanished. In an era when conversation with AI is daily life, the boundaries between thought, language, and technology are melting.

That I picked up poetry and philosophy after using AI, that I started spending time on foreign languages — maybe this is the byproduct of learning a new “language”: a sensitivity to language itself, awakened in passing. The new linguistic practice called prompting has led to an inquiry into the nature of language.

8. The Skills Humans Should Cultivate in the AI Era

Pulling the threads together, the topography of what’s demanded of humans in the AI era begins to take shape.

First, metacognition. The capacity to observe, structure, and evaluate your own thinking. To write a good prompt to AI, you first have to know what you want. The ability to ask yourself, “What problem am I actually trying to solve?” This is the most basic competency of the AI era.

Second, the art of the question. AI can produce the answer. But only humans can pose the question. What question you ask decides what answer you get. Good questions are rarer than good answers, and so they’re more valuable. In the AI era, the source of value is not the answer but the question.

Third, humanistic sensibility. Poetry, philosophy, language, history — in the AI era these aren’t “soft skills” but core skills. When AI generates average code, average writing, and average analysis in seconds, human differentiation comes from depths AI cannot reach. Those depths aren’t reachable by reading only technical documentation.

Scale AI hiring poets, developers studying philosophy, AI researchers rediscovering the humanities — all of it points the same way. The more humanlike machines become, the more humanlike humans must be.

In retrospect, what Claude Code taught me wasn’t coding technique. It was the shape of my own thinking.

Ask a vague question and get a vague answer; ask a precise question and get a precise one. This simple feedback loop changed me. My sentences got longer, the time I spent organizing thought grew, a sensitivity to language woke up, and my hand reached for poetry, philosophy, and foreign languages.

If AI is a mirror, what matters isn’t the resolution of the mirror but the depth of the person standing in front of it. The mirror is getting sharper. Are we?

Borrowing Shannon Vallor’s line: the AI mirror only shows who we are now; it cannot tell us who we might become. That possibility lies outside the mirror. In the step that closes the prompt window and walks to the bookstore, in the slow time spent reading a single line of a poem, a future self that AI cannot show is taking shape.

The most important question of the AI era is not “what should I have AI do?” It is “who will I become in front of AI?”

References

Shannon Vallor, The AI Mirror (2024) — A philosophical framework that treats AI as a mirror of human self-recognition
Andrej Karpathy — “The hottest new programming language is English” (2023)
ACM CHI 2024 — Paper on metacognition and ambiguity tolerance among GenAI users
Frontiers in Artificial Intelligence (2025) — The “Cognitive Mirror” framework paper
Harvard Gazette (Nov 2025) — A warning on “cognitive atrophy” in the AI era
Anthropic (Jan 2026) — Study showing a 17% drop in code comprehension among developers using AI coding tools
MIT Media Lab — Brain-monitoring research on AI-assisted writing and memory loss
Pew Research Center (Jun 2025) — American perceptions of AI’s effect on creative thinking
Kenneth Iverson, “Notation as a Tool of Thought” (1979 Turing Award lecture)
Luciano Floridi — The concepts of “semantic pareidolia” and “onlife”
Simon Willison — The principle of preserving one’s own voice when using AI
Ethan Mollick — “Just do stuff with AI for about 10 hours”
Mitchell Hashimoto — Work-decomposition skill in the AI era
Darius Foroux — “I want to out-human AI writing”