TL;DR
Get the little things that make your day delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A November 2026 blog post by computer scientist Murat Buffalo argues that writing is not just a way to record ideas: it helps people develop, test and remember them. He says AI-generated prose can look like evidence of thought without requiring the writer to think, and recommends producing an unpolished draft before editing or turning to a model.
Computer scientist Murat Buffalo argues in a November 2026 blog post that writing helps people think through ideas, not merely record them, and says relying on AI-generated prose too early can replace some of that work. His account combines personal experience with examples from research and computer science; it is an argument, not a reported study measuring writing’s effects.
Buffalo traces his view to middle school, when he struggled to retain long passages while reading. He began taking notes and making mind maps, then found that arranging information into branches helped him work out how ideas related. In hindsight, he writes, those pages served less as storage than as a way to process what he read.
He later connected that experience to advice attributed to computer scientist Manuel Blum. Blum compares a finite automaton, which has limited computational capability, with a Turing machine, which can use paper and pencil as external memory. Buffalo uses the analogy to explain how writing can let a person hold and manipulate ideas beyond what they can keep in their head. It is an explanatory analogy, not a claim that writing literally changes a person’s computational class.
Buffalo says drafting research reviews has shown him gaps in his understanding: he may feel he understands a paper, then discover while writing that he has grasped only part of it. He also recounts starting a 2018 blog post about similarities between Paxos and Nakamoto consensus without beginning with that conclusion. As he wrote, the connection became clearer, and he says the post later became a seed for a research paper.
Why Drafting Can Change the Idea
The practical point is that writing can make a person’s reasoning visible enough to examine. A draft gives the writer something concrete to question, reorganize and connect with other knowledge. Buffalo describes this as external memory: notes and blog posts let him return to ideas and rebuild his understanding, rather than relying on what he can retain unaided.
His argument also speaks to how people use large language models. A model can produce fluent text, but fluency alone does not show that the user has worked through the issue. Buffalo warns that opening a chat before forming an initial view may crowd out a person’s own half-formed intuition. That concern matters in education, research and workplace writing, where a polished document can be mistaken for evidence of careful reasoning.
The post does not establish that AI always weakens thought or that people should avoid it. Its narrower claim is that writing without a model can serve a distinct purpose: developing an idea before asking a tool to summarize, critique or extend it. Readers can treat that as a practical suggestion, while recognizing that the post offers experience and reasoning rather than controlled evidence.
Popular right now
Vitamix Propel Series 510 Professional-Grade Blender, Black
KitchenAid Classic Series 4.5-Quart Tilt-Head Stand Mixer, Onyx Black
Roborock RockNeo Q110H Wire-Free Robot Lawn Mower
As an affiliate, we earn on qualifying purchases.
From Mind Maps to AI Drafts
Buffalo’s account links several stages in his own working life: note-taking as a student, writing during doctoral work, reviewing research papers and keeping a blog he later consults. The common thread is using text to work through material that would be difficult to hold entirely in memory. He likens this process to out-of-core execution, a computing approach that uses external storage when information cannot fit in working memory.
He also invokes the six-page narrative memo associated with Amazon’s internal writing culture. Buffalo cites Jeff Bezos’s claim that writing such a memo requires clear thinking, then argues that this is less reliable in the age of generative AI: a person can now produce several pages without doing the underlying reasoning. That is Buffalo’s interpretation of the format’s limits, not evidence about Amazon’s current practices or a measured assessment of AI-written memos.
For his own method, Buffalo recommends what he calls “messing up”: writing freely, without stopping to correct errors or produce polished sentences. The aim is a rough initial draft that can be examined and improved. He compares a working notes file to a research lab notebook, where tangents and incomplete thoughts can be recorded before anyone tries to make them presentable.
“Without writing, you are reduced to a finite automaton. With writing you have the extraordinary power of a Turing machine.”
— Manuel Blum, as quoted by Buffalo
What the Post Does Not Establish
The post does not present an experiment comparing people who write before using AI with those who ask a model first. It supplies no data on learning, memory, creativity or work quality, so the scale and reliability of the effects Buffalo describes remain unmeasured.
Buffalo also does not define a precise boundary between useful AI assistance and a substitute for a person’s own reasoning. Models’ capabilities vary and continue to change; his view that they cannot be trusted to make a creative leap is his assessment, not a settled finding about all systems or future models. The post likewise offers no evidence that AI-generated memos have broadly displaced human reasoning in workplaces.
How to Apply the Drafting Advice
Buffalo’s suggested next step is straightforward: start a notes file, write freely while working through a problem, and postpone polishing until there is material to assess. The rough draft can then be used to identify weak reasoning, unresolved questions and possible connections. This is advice from the post, not a formal method with tested results.
Readers who use AI can also choose when to bring it in: for example, after writing an initial explanation or list of questions, rather than before forming any view. The post sets out no follow-up study, product change or institutional policy. For now, its contribution is a clear position in an active debate over whether fluent AI output supports thinking or can stand in for it.
Key Questions
What is the main claim of the blog post?
Writing can help develop ideas, not just preserve them. Buffalo says drafting gives people a way to examine, connect and remember thoughts.
Does the post show that writing improves thinking in a study?
No. It draws on Buffalo’s personal experience, examples from research work and computer-science analogies. It does not report a controlled study or quantify an effect.
What does Buffalo recommend for getting started?
He recommends writing a rough first draft without pausing to correct every error. The goal is to create material that can be reviewed and improved later.
Does Buffalo say people should not use AI?
No. He warns against handing over the early thinking process by opening a chat immediately, but the post does not call for avoiding AI altogether or lay out a complete AI-use policy.
Source: muratbuffalo.blogspot.com
