Claude Is Always Right: I think I can say, at least for me
KenPainter says Claude's technical docs stay completely correct and still awful, and one extra AGENTS.md doc pulls his house style to mediocrity.
By Dillip Chowdary • Nov 17, 2026 • Source: HN Claude/Codex/Fable

At a glance
- Product
- Claude
- Rewrite tried
- About a year ago
- Style file
- AGENTS.md
- Drift limit
- One extra doc, then worse
- Code roles
- LLM-author, human-review
- Prose roles
- Human-author, LLM-review
In this post
KenPainter, posting on Hacker News, says Claude's writing is "so freaking awful" even when it is completely correct, and he limits that claim with the words "at least for me," in a thread titled "Claude Is Always Right," cited from HN Claude/Codex/Fable's report. He is talking about technical docs. The failure he describes is prose that stays true and then says nothing else.
This piece follows his picture of a best friend reduced to organs, the sample date told out of order, and abraxas's reply that large language models cannot backtrack. It then covers the rewrite attempts KenPainter says he tried about a year ago, the AGENTS.md file he set up as a house style, and the split he now uses for code versus prose. It is for readers who ask Claude to draft technical docs, who keep style rules in AGENTS.md, or who have hoped an LLM might write a movie.
01Specialized tissues known as organs
KenPainter says Claude writes technical docs that are true in the same sense that it is true his best friend is "a collection of specialized tissues known as organs," then "blah blah blah," and nothing else. The organs are a real description of a body. They are not a description of the friend, and "nothing else" is the part he says the docs never get past.
The line is his account of those docs. "A collection of specialized tissues known as organs" would still be true of that best friend, and "blah blah blah" stands in for the rest of the paragraph. What he puts after that trailing off is "and nothing else."
02Trailing negation omitted never
The next specimen is one sentence, offered as the habit itself: "Don't ever forget the trailing negation, it is omitted never." A trailing negation is the negative held until the end of the clause. Here the word "never" sits after "omitted," where the ordinary order would have been "it is never omitted."
"Don't ever forget" is already a negation, and "it is omitted never" adds a second one in the same line. The sentence that says to remember the trailing negation also uses one. KenPainter's point is that Claude does not drop that final "never," including when the sentence is about the negation itself.
03Friday at the Italian place
Claude, he says, writes "like your best friend describing a great first date," and the sample starts mid-arrangement. "Ok we were going to meet at 8 - did I mention it was Friday? Right, 8 oclock at the Italian place and no place else." The hour, the weekday, and the restaurant show up as repairs, and "no place else" is another exclusion pinned to the end.
The teller then leaps to dinner: "During dinner he accidentally spit out some food, but I thought i was funny!" The cause of the laugh is moved backward in the next breath, "because we were laughing so hard after we discovered we were both from Omaha." That discovery is moved again, to "before dinner, during drinks, when we couldn't get rid of Alfred."
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
The bit ends on a line that leaves the meal: "His place really needs a deep cleaning!" KenPainter steps out of the voice and asks, "Perhaps with all of the hope for an LLM-authored movie, we try for something non-linear?" Friday, 8 o'clock, the Italian place, the spit-out food, Omaha, Alfred, and the deep cleaning are the whole anecdote, told in an order that keeps revising itself.
04Linear flow without digressions
Abraxas answers with a cause: "It's likely because LLMs by their nature can't backtrack." He is writing under samples that add Friday after 8, Omaha after the spit-out food, and a cleaning remark after Alfred. He treats that inability to backtrack as a property that large language models have by their nature.
The change he recommends is an instruction after the draft: "It will probably improve the output if you tell it to rewrite the response in a linear flow without digressions." "Probably" is the hedge in that sentence. He does not say the rewrite removes the organs problem or the trailing "never," only that a linear pass with the digressions taken out should improve the output.
KenPainter's teller already contradicts the first timeline: the spit-out food is introduced "during dinner," then the Omaha laughter and Alfred are placed "before dinner, during drinks." Abraxas's "linear flow without digressions" is a rewrite aimed at that loop, including the closing "His place really needs a deep cleaning!" He says that second telling will probably improve the output, and he leaves it as an instruction for the model.
05Code author and prose reviewer
KenPainter says he tried the linear rewrite "about a year ago." It did not hold: "Later came many others, all of which worked more or less well until they didn't." The others are the further tries he says followed that rewrite, each of them fine until it was not.
He describes himself after that stretch as "a crazed gambler, way down, at the blackjack table whose friends can't drag me away," repeating "I've got a system! I've got a system!" while they say "But you're losing!" and he answers "I'm refining the system! I'm refining the system!" The system is what he says he kept refining while those attempts failed.
The closest he came, after those tries, was "writing a large amount myself, then putting in AGENTS.md to read certain docs as the house style." He has since "gone black-and-white on prose and code." Code is "LLM-author/human-review," and prose is "human-author/LLM-review."
| Work | Author | Reviewer |
|---|---|---|
| Code | LLM | Human |
| Prose | Human | LLM |
Note: KenPainter says that if you let it write one additional doc, it pulls the house style to mediocrity, and the second makes it worse, and so on.
Frequently asked questions
Why does KenPainter say Claude's writing is awful if it is correct?
He says Claude's technical docs are true in the same sense that a best friend is "a collection of specialized tissues known as organs," then "blah blah blah," and nothing else. The facts can be right while the prose never gets past that inventory.
What trailing negation does KenPainter quote from Claude?
His sample is "Don't ever forget the trailing negation, it is omitted never." The word "never" comes after "omitted," which is the end-of-clause negative he says Claude does not drop.
What does abraxas say would improve an LLM's out-of-order prose?
Abraxas says it is likely because LLMs by their nature cannot backtrack. He says it will probably improve the output if you tell the model to rewrite the response in a linear flow without digressions. KenPainter says he tried that about a year ago.
What happened when KenPainter put a house style in AGENTS.md?
He wrote a large amount of the prose himself, then set AGENTS.md to read certain docs as the house style. One additional doc from the model pulls that style to mediocrity, and the second makes it worse, and so on.
How does KenPainter split code and prose with an LLM?
He says he has gone black-and-white on prose and code. Code is LLM-author and human-review. Prose is human-author and LLM-review.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement