When the prose feels off, the thinking is off: Paul Graham's insight on writing and cognition

Paul Graham published a short but dense essay this week — "Good Writing," dated May 2025 — that looks, on its surface, like a meditation on craft. Read it again, and it's actually a theory of how rigorous thinking works. For founders who write strategy memos, product briefs, investor updates, and pitch narratives, the argument has direct operational consequences. 1

The core claim: writing quality and idea quality are the same signal

Graham opens with what he calls "the most exciting kind of idea: one that seems both preposterous and true." His thesis: writing that sounds good is more likely to be right.

Not right in a narrow "no typos" sense. By "right" he means developed well — drawing the conclusions that matter most, exploring each to the right depth. The two attributes founders might instinctively treat as orthogonal — how a document reads versus whether its logic holds — turn out to share a root.

The mechanism he gives is precise. When you force yourself to rewrite an awkward passage, you cannot do it in a way that makes the ideas less true — "you couldn't bear it, any more than gravity could bear things floating upward." So any change to the language has to improve the ideas. He uses the analogy of shaking a bin of objects: each shake is arbitrary, but gravity ensures that random perturbations only ever produce tighter packing. Rewriting for sound is that shake. The ideas are the objects rearranging themselves.

"So yes, the two senses of good writing are connected in at least two ways. Trying to make writing sound good makes you fix mistakes unconsciously, and also helps you fix them consciously; it shakes the bin of ideas, and also makes mistakes easier to see." — Paul Graham, Good Writing (May 2025) 1

For founders building in 2025: Every time you revise a product spec because it "doesn't read cleanly," you are not just polishing language. You are doing logic work. The prose discomfort is a symptom pointing to a structural problem in the underlying reasoning.

The first-reader principle and why it matters for internal docs

The second mechanism Graham identifies is subtler. Good writing is easier to read. That helps the writer because — and this is the sentence worth stopping on — the writer is the first reader.

He describes rereading some paragraphs 50 or 100 times during the drafting of a single essay, asking whether anything "catches." The easier the writing flows, the easier it is to notice when something snags. Difficulty in reading is camouflage for difficulty in thinking.

Person writing in a notebook, deep in thought — A founder's strategy memo is a thinking instrument before it is a communication artifact. 2

This flips how most founders treat internal writing. Internal docs — design docs, strategy memos, Q2 roadmaps — routinely get a lower editorial bar than external-facing work. The reasoning is: it's internal, only the team reads it, good enough is fine. Graham's argument reverses the priority. Internal writing is exactly where the first-reader benefit is most valuable. The team document you write to "capture the thinking" is not downstream of the thinking; it is the thinking in progress.

Document type	Most founders treat it as	What Graham's argument implies
Investor update	Communication output	Logic checkpoint — does the narrative cohere?
Product spec	Reference artifact	Cognitive instrument — bad flow = missing logic
Strategy memo	Internal alignment tool	First-reader test — if it's hard to read, the strategy has a hole
Pitch narrative	Sales material	Consistency test — beautiful but false is internally consistent, not externally valid

Rhythm as a validity signal

The third mechanism is the most counterintuitive. Graham argues that good prose rhythm is not the regular beat of verse; it's the natural rhythm of the thought itself. An essay "is a cleaned-up train of thought," and a train of thought has shape. When the rhythm is wrong, it usually means the thought's shape is wrong.

He makes this practical in a parenthetical remark: "Often I don't even distinguish between the two problems. I just think Ugh, this doesn't sound right; what do I mean to say here?"

The italicized sequence is the actionable part. The prose instinct — the feeling that a sentence is off — is a legitimate trigger to re-examine the idea. Not to rephrase it. To reconsider it.

For early-stage AI founders specifically: you write in a domain where the underlying technical claims are genuinely hard to communicate. The temptation is to attribute awkward prose to subject-matter complexity. Graham's framework suggests a different prior: if the section about your model's inference architecture reads oddly, consider that the section's logic may be strained, not just the vocabulary.

paulgraham.com

Good Writing — Paul Graham

There are two senses in which writing can be good: it can sound good, and the ideas can be right. Paul Graham argues these two kinds of good are not unrelated — and the implications for how founders think are significant.

링크 미리보기를 불러오는 중…

Where the mechanism breaks down — and why founders need to know the boundary

Graham is careful with scope, and the boundary is commercially significant. The connection between sounding good and being right holds only "when you're writing to develop ideas." It does not apply when you already have the ideas and are writing to describe them afterward — a paper about an experiment you've run, a textbook summarizing others' findings.

The founder-relevant corollary: a polished Series A pitch deck is not evidence that your business logic is sound. By the time you're writing a deck, you're largely in description mode, not idea-development mode. The deck that reads beautifully may be one you've written after the thinking was done — and that means it lacks the self-correcting benefit Graham is describing. Beautiful pitch decks can and do describe bad businesses coherently.

The practical protocol is to use writing-as-thinking upstream, before the pitch. The weekly internal memo, the rough thinking doc, the "what are we actually trying to do" note — these are the documents where the first-reader mechanism operates. The deck is downstream. Use it as a check, not as the place where the logic develops.

The liar's corollary: internal consistency is not truth

Graham anticipates the obvious objection: liars write beautiful prose. He has a direct answer: to write something beautiful and false, you begin by "making yourself almost believe it." A smooth-tongued liar is presenting a perfectly formed train of thought — the difference from an honest writer is only where the argument attaches to reality.

So the sharper version of his thesis is this: better-sounding writing is more likely to be internally consistent. If the writer is honest, internal consistency and truth converge. If not, they don't.

This matters for a specific failure mode that's common among technically capable AI founders: narrative self-seduction. The ability to write clearly and build internally consistent arguments is, in early-stage startups, often stronger than the ability to verify those arguments against external reality. A well-written strategy that "makes sense on paper" can survive many internal reviews precisely because everyone is evaluating internal consistency, not external validity.

Graham's corollary is a useful prompt: Is my narrative beautiful because the logic holds, or because I've started to half-believe the premises? The countermeasure is to make the premises of any key narrative explicit, then test them against external data points independently — not by asking "does this still make sense," but by asking "what would have to be false for this to be wrong."

One practical implication for how you allocate writing time

The full argument implies a reallocation that goes against most founders' habits. If writing that sounds good is a mechanism for getting thinking right, then the documents that deserve the most editorial effort are not the ones most people will read — they are the ones where the thinking is still forming.

That means:

The rough exploration doc where you're deciding whether to build feature X deserves revision energy.
The one-pager you circulate internally before making a major hire deserves revision energy.
The "what's our positioning, actually" document deserves revision energy.

The Series B deck, the press release, the case study that will be on the website — these already have the thinking baked in by the time you write them. Polishing them is communication work, not cognition work. Both matter. But the first-reader benefit runs only in one direction.

Graham has returned to the question of how founders should think and write repeatedly. His 2024 essay "Founder Mode" — a direct argument that founders need to operate differently from professional managers — is worth reading alongside "Good Writing" as a pair. Together they build a case that a founder's most durable advantage is the quality of their thinking, and that writing is one of the clearest instruments for developing it.

paulgraham.com

Founder Mode — Paul Graham

The essay that sparked widespread debate about whether founders should adopt manager-style delegation or stay deeply involved in their companies' decisions.

링크 미리보기를 불러오는 중…

Graham's "Good Writing" essay is two pages. It is worth reading slowly, then rereading the section on rhythm. Published May 2025. 1

When the prose feels off, the thinking is off: Paul Graham's insight on writing and cognition

The core claim: writing quality and idea quality are the same signal

The first-reader principle and why it matters for internal docs

Rhythm as a validity signal

Good Writing — Paul Graham

Where the mechanism breaks down — and why founders need to know the boundary

The liar's corollary: internal consistency is not truth

One practical implication for how you allocate writing time

Founder Mode — Paul Graham

참고 출처

Good Writing — Paul Graham

Founder Mode — Paul Graham