It's funny that this post came out right after I set up my first real blog. This timing allowed me to prompt an LLM in the same conversation I'd been having about setup.

Claude> Talk to me about https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/

Claude answered with some mildly interesting disagreement and agreement, and then said the following:

What's interesting reading it against this conversation: what we've actually been doing tracks closer to the article's model than most "AI writing help" — I never drafted your README prose. You wrote every sentence of it yourself across several passes; my role was pushing back on the reasoning (the CMS claim, the "consistency" framing, the standards-vs-ergonomics conflation) and then executing the resulting decisions. That's closer to "copyeditor surfacing problems" than "ghostwriter," which is presumably the mode you'd want for the blog itself too, given why you built it this way in the first place.

So, to be clear, I absolutely agree with the main thesis. And because I find the main thesis interesting and relevant, I want to argue with some of the details.

Yay for LLM copyediting and boo for LLM ghostwriting

I like dry humor and I'm often silly, but you'll notice I didn't put any tongue-in-cheek language in here that mocks 2026 LLM writing style. Because to a lot of people today, it's not funny. People are sick of seeing that particular voice.

But it isn't the detectable patterns themselves that irritate people. They're irritating because they're associated with soulless writing. All an LLM does is string words together in a believable pattern. They don't have any goal in their writing that hasn't already been articulated earlier in the token sequence. An LLM is an amazingly useful collection of information about how words relate to other words, but it has no idea what the words mean. And that comes through in the rambling, useless verbiage that frustrates readers trying to understand the "writer's" intent. Soo boo for LLM ghostwriting.

Yay for LLM copyediting. Having a model, trained on a huge volume of ideas that humanity has articulated, at your beck and call any time of day or night is a huge win for writing.

It’s tricky to write about writing?

I disagree with "It’s tricky to write about writing. It comes across as a brag; you’re implying that you write well." The fear being expressed is that you'll be seen as having opinions on what makes good writing when you yourself aren't producing good writing. So what? We sit in our armchairs and express opinions about how professional athletes should play, without being embarrassed that we're not able to execute the plays we're recommending. My opinions about what is good writing don't have to be self-referential. And suppose you are bragging? So what? Strong opinions are fun to read. For example, when I was in school my favorite teachers were those who were convinced that their own subject was the most important knowledge one human could impart to another.

Writing about writing is so natural that screenwriters can't seem to not do it. Every movie and TV show has a character who writes, or who is obsessed with books. I see such themes so often that I kind of wish it was tricky to write about writing.

You may not use a single word an LLM suggests to you?

I would be sympathetic if he'd written not to put any two words together that came from an LLM. But an LLM is an awesome place to find a single word. If you use a thesaurus to find a word, you have to come up with a synonym first. A large language model lets you describe the word you're looking for.

Toward the end of that section he shifts toward arguing against LLM-generated phrases. He probably should have used an LLM copywriter to detect that inconsistency.

Avoid encouragement?

I think it's going to far to forbid the model from encouragement. An LLM is an amalgamation of the thoughts of lots of humans, and encouragement from it is just a reminder that somebody, somewhere might get something good out of some of your writing. If LLM encouragement is enough to get you to stop examining and fixing what you've written, you're not going to produce good writing anyway.

OK, maybe ghostwrite the closing

Earlier, I said that LLMs "don't have any goal in their writing that hasn't already been articulated earlier in the token sequence". After I wrote that, it occurred to me that an essay closing doesn't have to introduce any new ideas, so you could actually have an LLM ghostwrite your closing and it might do a good job. Is that what I've done here? I find it amusing not to tell you.

The fastest and easiest way to configure this theme is the Lume init command, which you can also copy easily from the Simple Blog theme page. Running:

deno run -A https://lume.land/init.ts --theme=simple-blog

will create a new project with Simple Blog configured. Edit the _data.yml file in your blog root folder with your data to customize the site title, description, and metadata.

Posts must be saved in the posts folder. For example, posts/my-first-posts.md.

Install as a remote theme

To add the theme to an existing Lume project, import it in your _config.ts file as a remote module. Update it by changing the version number in the import URL:

import lume from "lume/mod.ts";
import blog from "https://deno.land/x/lume_theme_simple_blog@v0.15.6/mod.ts";

const site = lume();

site.use(blog());

export default site;

Copy the _data.yml file to your blog root folder and edit it with your data.

Customization