Writing as Thinking: Reassessing the Cognitive Value of Writing in the AI Era

Writing Is Not Just Output—It's an Engine for Thought

Now that AI can generate articles, emails, and reports in seconds, it seems reasonable to ask: do we still need to write things ourselves? Murat Buffalo offers a compelling response in his blog post Writing as a thinking tool—the core value of writing isn't producing text, but its function as a thinking tool. The discussion, which earned 180 points and 59 comments on Hacker News, touches on a hidden anxiety among knowledge workers in the wave of intelligentization: when generation becomes cheap, is thinking also depreciating?

Why Writing Can Leverage Deeper Thinking

Linear Expression Forces Logical Organization

Writing requires us to transform web-like, jumpy thoughts into a linear sequence of words. This process itself forces the author to clarify concepts, establish causality, and discover logical gaps. Often, we believe we've thought through a problem until we put it into writing and find the argument full of holes. Murat Buffalo's observation is that writing is not a record of thinking but the way thinking happens—sentences shape ideas even as they are formed.

This is especially evident in technical and academic writing. Design documents, paper drafts, architecture notes—the process of producing these texts is often accompanied by the refinement of decisions. At the moment of writing, vague intuitions are forced to undergo logical scrutiny.

The "Cognitive Friction" Writing Provides

The biggest advantage of AI-generated content is fluency, and fluency may precisely dissolve the friction that thinking requires. When writing yourself, the discomfort of stalls, repetitions, and deletions is actually a signal that cognition is deepening. Buffalo suggests in his article that skipping this friction is equivalent to skipping thinking itself. Readers in the HN comments echoed this: outsourcing writing tasks to a model yields a seemingly complete text but loses an opportunity for self-clarification.

The concept of "cognitive friction" is borrowed from the field of human-computer interaction, originally referring to the user cognitive load caused by unsmooth operations in interface design. In the context of writing, it is used to describe the deliberately retained resistance in the thinking process: when you can't find the right word, the logical chain breaks, or the argument is self-questioned, these discomforts are not obstacles to be eliminated but triggers for deeper processing. The "desirable difficulties" theory in cognitive psychology resonates with this—moderate difficulties in learning and thinking actually promote long-term memory and understanding. The fluent text generated by AI precisely eliminates this resistance, allowing writers to skip conceptual analysis and logical verification and obtain a superficially smooth product lacking internal connections. Therefore, retaining friction in writing is essentially retaining a window for self-correction and meaning construction.

In the AI Era, the Role of Writing Becomes Clearer

Distinguishing "Generation" from "Thinking"

AI excels at generation—predicting the next word based on context. The value of human writing lies in judgment: does this argument hold up? Is this structure reasonable? Is this wording accurate? When the cost of generation approaches zero, judgment and critical thinking become scarce capabilities. Writing, as a way to train judgment, becomes more important, not less.

This does not mean rejecting AI assistance. A more pragmatic stance is: use AI for low-cognitive-load formatting work, and reserve writing energy for parts that require real thinking—problem definition, argument structure, critical reflection. The key is to retain the "write it yourself first" step in the workflow.

AI's "generation" is based on the next-word prediction mechanism of large language models (LLMs), which produce coherent text according to contextual probability distributions but do not build real beliefs or causal models about arguments. Human "thinking" involves concept manipulation in working memory, hypothesis testing, and metacognitive monitoring. The key difference between the two is: generation pursues local coherence and statistical plausibility of text, while thinking pursues the truth value of propositions, the validity of arguments, and the appropriateness of goals. When writers mistake generation for thinking, they easily fall into the "fluency illusion"—the text reads smoothly but has not been tested by logic and evidence. Distinguishing the two is not to deny the auxiliary value of AI, but to remind writers: judgment, skepticism, and self-criticism cannot be replaced by statistical prediction; these abilities need to be honed repeatedly in the practice of writing by hand.

Writing as Practice Against Intellectual Laziness

In an environment of information overload, passively consuming opinions is far easier than actively constructing them. Writing is a forced active construction: you must decide what to say, in what order, and why. This structured thinking training cannot be replaced by watching any number of short videos or reading any number of summaries. Buffalo's article is essentially reminding knowledge workers not to outsource the responsibility of thinking along with the writing task.

Implications for Technical Professionals

For engineers and product people, the value of writing is often underestimated. Writing an RFC, a postmortem, or a design trade-off explanation—these actions are not only communication tools but also opportunities to train systems thinking. Buffalo's blog post and the HN discussion together point to a conclusion: in an era where AI tools are readily available, actively choosing to write personally at critical moments is a clear-headed cognitive investment.

Writing will not lose its meaning because of AI; its meaning is simply being repositioned—from "producing documents" to "producing clarity."

Frequently Asked Questions

Can writing and AI assistance coexist? Yes, but it is recommended to do it in stages: first complete the thinking-intensive writing independently, then use AI for polishing and format optimization, avoiding letting generative models replace your judgment.

Why is it said that writing can improve thinking quality? Because writing forces vague ideas to be transformed into clear propositions, exposing logical leaps and conceptual confusion—this process itself is the deepening of thinking.

One-line Summary

Writing is how thinking happens; in the AI era, its value for training judgment only increases.

Paragraph Summary

The article centers on Murat Buffalo's Writing as a thinking tool and the Hacker News discussion, pointing out that the core value of writing is not producing text but its function as a thinking tool. Writing transforms web-like thinking into linear expression, forcing logical organization and exposing argument gaps; the stalls and repetitions in personal writing constitute "cognitive friction," a signal of deepening thinking, while the fluency of AI generation may dissolve this friction. When the cost of generation approaches zero, judgment and critical thinking become scarce capabilities, and the importance of writing as a way to train judgment rises. The author advocates a pragmatic stance: use AI for low-cognitive-load formatting work, but retain the "write it yourself first" step, avoiding outsourcing the responsibility of thinking along with the writing task.