Quick Answer: Writing to think means using the act of writing to clarify, test, and develop ideas. Writing forces you to externalize half-formed thoughts, which exposes gaps and creates connections your head alone would miss. Computer scientist Manuel Blum framed it as the difference between a finite automaton and a Turing machine: paper and pencil add a huge amount of computational power to your reasoning.

Most people think writing is how you report what you already know. That's backwards. In a recent post on his blog, systems researcher Murat Demirbas argues that writing is the place where thinking actually happens. He describes reading a page with nothing sticking, then discovering that drawing notes was how he finally made sense of it. Once you see the mechanism, you'll start treating your notebook differently.

The Turing Machine Argument

Blum's advice to graduate students makes a crisp technical point. A finite automaton can add but not multiply. A Turing machine can compute any computable function. The only difference between them is that the Turing machine has a tape it can write on.

Take that seriously and the implication is uncomfortable. If you reason in your head alone, you're working with a bounded memory, like a finite automaton. Put a pencil and paper in front of you and you gain the ability to hold intermediate results, revisit them, and build on them. Demirbas calls this out-of-core execution, borrowing the term from algorithms that use external memory when the data won't fit in RAM.

The analogy is a metaphor, not a theorem. Nobody has proven that human cognition maps neatly onto complexity classes. But the working intuition holds up: the paper isn't a record of your thinking, it's a component of it.

Reviewing a Paper Changes What You Understand

The clearest example in the post is peer review. Demirbas reads a paper, feels he understands it, then sits down to write the review and discovers he grasped maybe 60% of it. That number is his personal estimate, not a study, but the experience will sound familiar to anyone who has reviewed or taught.

Writing the review forces you to state claims precisely, check them against the text, and say what the paper's contribution is in your own words. The gaps show up as sentences you can't finish. Here's where it gets interesting: sometimes the writing surfaces connections the authors never noticed.

Demirbas describes this happening with a 2018 blog post comparing the anatomical similarities between Paxos and Nakamoto consensus. He didn't start with that thesis. The connections clarified as he wrote, and the post became the seed for a research paper. That's freewriting in its useful form: you write to find out what you think.

Why the Blog Works as External Memory

Writing also helps you remember. Demirbas says he rereads his own posts constantly, and that reconstructing the understanding is fast because the original writing strained his thinking. Rereading reactivates the same mental pathways.

Leslie Lamport put it bluntly: if you think without writing, you only think you're thinking. The thoughts you never write down tend not to stick, because they were never fully formed.

That said, there's a real catch. Rereading your own notes works only if the notes capture reasoning, not just conclusions. A file of bullet fragments like "use quorum, check later" tells you nothing six months on. Write the why, the assumption, and the counterexample you considered and rejected.

The Five Rules: Draft, Cut, Simplify, Ignore the Audience, Schedule

Demirbas's process is worth copying because each rule solves a specific failure mode.

  1. Mess up first. Produce in bulk without criticizing. Keep a notes file as a lab notebook, follow tangents, and skip the typo fixing. Ideas are fragile early on, so protect them from your own inner critic.
  2. Tidy up as a harsh reviewer. Pretend someone else wrote the draft and cut ruthlessly. Move good sentences you can't bear to delete into a junk buffer at the bottom. You won't go back to them.
  3. Keep it plain. He's a fan of Strunk and White's The Elements of Style, and of Kurt Vonnegut's plain prose. Complexity often hides sloppy thinking or lets you avoid taking a stance. If you can't say it plainly, you probably haven't finished thinking.
  4. Write for yourself. External rewards are delayed and random. Demirbas has blogged since 2011 and only later found that readers were showing up at conferences. The lesson is that the motivation has to come from the writing itself.
  5. Put it on the schedule. Waiting for inspiration rarely works. A regular posting cadence keeps you in what he calls a "noticing frame of mind," and he often finds the topic he dismissed becomes the most interesting one once he starts.

McPhee's observation that you should write a piece three or four times, never once, ties these together. Multiple drafts take the pressure off. You don't need to be brilliant on the first pass.

Does Writing Still Matter When an LLM Can Draft for You?

This is the question the post handles most sharply. Yes, a model has read the literature and knows the standard framing. But it doesn't know your half-formed intuition about why the standard framing fails for your specific case. Demirbas doesn't trust current LLMs to make that creative leap, and he suspects it would take another breakthrough.

The risk is subtle. Amazon's six-page narrative memo was praised because, as Jeff Bezos put it, you can't write one without clear thinking. Demirbas's point is that this no longer holds. A model can produce six polished pages with no thinking behind them, and the pages still look like evidence that thinking happened.

So the advantage shifts. If you open a chat before you've struggled with the problem on paper, you skip the step where understanding gets built. Writing it yourself becomes a comparative advantage. Here's the uncomfortable part: the people who keep doing the work will likely understand more than the people who outsource it.

There's a fair counterpoint. Some people genuinely use LLMs to argue with, stress-test drafts, or summarize prior work, and that can sharpen thinking. The line is whether you wrote your own position before you asked for the machine's opinion.

Comparison: Writing Modes and What Each Gets You

ModeWhat you produceThinking it developsMain risk
Lab notebook (freewriting)Messy, unedited notesExploration, connectionsNever revisited
Paper reviewStructured critiquePrecise claim checkingGaps hidden by polish
Blog postFinished argumentClarity, memory retentionWriting for applause
LLM-drafted textPolished proseLittle, if you don't write firstLooks like thinking without being it

Frequently Asked Questions

What does writing to think mean?

Writing to think means using writing as a tool to work out ideas, not just to record them. You put half-formed thoughts on paper, which exposes gaps and makes you state assumptions clearly. Demirbas argues the paper becomes part of your reasoning, extending what you can hold in mind at once.

How do I start writing to think if I don't know what I believe yet?

Start with a messy first draft and don't edit it. Write the question, your current guess, and what would change your mind. Then reread and mark the sentences you can't finish. Those unfinished sentences show you where your thinking needs work.

Why does writing help me remember things better than reading?

Writing forces you to reconstruct an idea in your own words, which is harder than recognizing it on a page. Rereading your own notes later reactivates that effort. Demirbas says his blog works as external memory because the posts were written with real strain, so rereading them quickly rebuilds the understanding.

Can I still use AI tools while practicing writing to think?

You can, but order matters. Write your own position first, then use a model to challenge it or check for missed literature. If you ask for the argument before you've formed one, you skip the step where the thinking happens and end up with text that looks like understanding.

Start Your Own Notes File This Week

Writing to think works because it slows you down while keeping you on the problem. Pick one paper, idea, or decision you're stuck on, and spend 20 minutes writing badly in a notes file before you open any chat window. Note what you found hard to explain, then come back the next day and see what changed. Pass this to a colleague who keeps saying they "understand" the material they can't explain.