Ezra Klein's Advice on Writing: The Case Against AI, and Why Writing Is the Last Step

Ezra Klein co-founded Vox, hosts one of the most popular podcasts in the world, and writes a column for the New York Times. He is one of the sharpest thinkers about how media, technology, and ideas interact.

On David Perell's "How I Write" podcast, Klein made the most articulate case against using AI for writing that anyone has offered. Not because AI is bad at writing. Because it is dangerous for the writer.

Writing Is the Last Step, Not the Core

Klein's central argument: writing is not where the real work happens. Writing is "selling the product." The manufacturing happened earlier. In the reading. In the reporting. In the interviews. In the hours spent with source documents.

Without that foundation, your writing is "rhetorically elegant recitation" of beliefs you already held. You sound good. You say nothing new.

He uses a baton-passing metaphor. Book writing is "constantly finding something to pass the baton to." You read a book on housing and that carries you for 1,600 words. You do a deep reporting trip and that carries you for 3,000 words. You conduct an interview and that gets a baton for another section. "You should very rarely be solving a problem through rhetoric." If you find yourself decorating weak ideas with strong sentences, the problem is not the prose. You have not done enough research.

The Danger of AI Summaries

Klein reads Congressional Budget Office reports. Full housing studies. Academic papers. Not summaries. Not bullet points. The full documents.

Why? Because knowledge is embodied. You do not download information like a file transfer. You grapple with it over time. "The time you spent with things is pretty important and a reason you get changed over time."

Having AI summarize a book defeats the purpose. You think you have read it. You have not. You have consumed a compressed version that left out everything your brain needed to make its own connections.

Klein's sharpest observation: "It's more dangerous to think you've read something than to acknowledge you haven't read it at all." The false confidence of AI-mediated knowledge is worse than honest ignorance.

He rejects the Sam Bankman-Fried view that every book should be a magazine article and every article should be a tweet. "No, it shouldn't be. In part because that actually doesn't contain the information. But also because it cannot change you." A magazine article changes you less than a book. A tweet changes you less than an article. Length is not waste. It is the vehicle for transformation.

Reading as Foundation

Klein reads for two to three hours daily in an undistracted state. Books, think tank papers, academic reports. This is not optional enrichment. It is the wellspring that enables everything else.

Extended reading creates what he describes as an associational mind state. Highlighting, connecting, cogitating. Your brain makes links between ideas that conversation and AI summaries cannot replicate. The slow immersion is the point.

He compares the process to travel. A place becomes alive to you once you have been there in a way that reading about it never achieves. Books work the same way. You have to spend time inside them for the knowledge to become part of how you think.

This maps to Klinkenborg's concept of "noticing." You cannot notice connections if you have not spent enough time with the material to see them. Speed is the enemy of insight.

AI Cannot Tell You Your Idea Is Wrong

This is Klein's most important point. ChatGPT will happily improve your prose. It will tighten sentences, suggest better word choices, restructure paragraphs. What it will never do is tell you that your core idea is fundamentally flawed.

"What ChatGPT will never tell you is that the problem with what you're doing is that it's the wrong thing entirely." It always answers the question of how to tweak and rewrite. It does not know when you should throw the whole thing out and start over.

That is exactly what writers need most. Not line edits. Not grammar fixes. The moment where someone says: this premise is wrong, this argument does not hold, you are deceiving yourself.

AI is a flattering mirror. A good editor is an honest one. The distinction matters.

The Nonlinear Creative Process

Klein describes the rhythm of book writing as deeply nonlinear. He would take five-day chunks at a cabin. For the first three or four days, he felt like he made no headway. The progress he made got thrown out. He thought he had wasted the week.

Then in the final two days, he would get so much done that the week was justified. "That structural work had to happen. That mental work had to happen. Those false starts had to happen."

This is the pattern that AI shortcuts destroy. If you let AI generate your first draft, you skip the struggle that precedes the breakthrough. The three wasted days were not wasted. They were the foundation for the two productive ones.

Writing as a Technology of Rigor

"Writing is a technology of rigor." In conversation, you can glide past weak points with rhetoric. In writing, the gaps are visible. You cannot hide a non-sequitur in a paragraph the way you can in a confident sentence spoken aloud.

Klein notices that writing also persuades the writer. "Recognizing as a writer, you always think you're convincing the audience. You're usually not convincing the audience of anything. You're convincing yourself." The process of building an argument narrows your thinking. You discard alternative explanations. You talk yourself into more certainty than the evidence warrants.

He wishes he were better at writing uncertainty. Most writers falsely resolve ambiguity because the medium demands a conclusion. But sometimes the honest answer is: I do not know yet. Very few writers can say that compellingly.

Say What You Are Saying

Klein had a sticker above his keyboard while writing Abundance: "Say what you're saying."

He is proud that the book is stripped down. Very little rhetorical flash. Declarative. Direct. "I am trying to say as directly as I can to you what I'm trying to say to you."

He warns against what he calls "jazz fingers" - writers using ornamentation to distract from weak substance. If the center of your argument has not been researched deeply enough, no amount of style will save it.

Trust Your Taste

Early in his career, Klein had good editorial judgment but did not trust it. He wished he had said no to more things at Vox. Not because his yes was wrong, but because his tolerance for mediocre work was too high.

Part of developing as a writer is learning to trust what you already sense. Artists often do their best work early, before the audience gets in their head. Klein's version of this: "I was much less willing to credit my own editorial instinct when I was younger."

This connects to Ira Glass's taste gap. You got into this work because you have good taste. Trust it, even when your current skill has not caught up.

Outwork, Do Not Outsmart

Klein tells young journalists that they can beat their elders simply by doing the reading. When he started in journalism, he actually read the Congressional Budget Office reports. They are not complicated. They are usually under 30 pages. But most journalists only read the executive summaries, if that.

"You could just beat people by outworking them, by reading the things that they had ignored, that they found too boring, that was not the part of the job they enjoyed."

This remains true in the AI era. Most people will use AI to skip the boring parts. The writer who actually reads the source documents, marks them up, and spends an hour thinking about them will see things the AI summary missed. The competitive advantage is not intelligence. It is willingness to do the work.

The Editor as Mirror

Klein believes the most important quality in an editor is taste. Not grammar knowledge. Not style guide expertise. Genuine judgment about what works and what does not.

His editor at the New York Times, Aaron Reica, provides something irreplaceable: trust. "The most important thing for a writer to feel about their editor is that their editor knows what is good and what isn't." Technical editing is secondary. What matters is that someone with genuine taste reviewed the piece before it went out.

Key Takeaways

• Writing is the last step. The real work is reading, reporting, and thinking.
• Pass the baton. Every section needs research, not rhetoric, at its core.
• AI summaries are dangerous. They create false confidence without real understanding.
• Read two to three hours daily. Slow immersion builds the connections AI cannot replicate.
• AI cannot tell you your idea is wrong. That is what you need most.
• Embrace the nonlinear process. The false starts are the foundation for breakthroughs.
• Writing is a technology of rigor. It exposes gaps that conversation hides.
• Be careful: writing persuades the writer first. Leave room for uncertainty.
• Say what you are saying. No jazz fingers. No rhetorical decoration.
• Trust your taste. Say no to more things.

Klein's argument is not anti-AI. It is anti-shortcut. Use AI for research acceleration and line editing. Do not use it to skip the thinking that makes your writing worth reading.

This post draws from Klein's appearance on How I Write and coverage in 3 Quarks Daily. Athens is an AI writing editor for writers who do their own thinking first.