Frank
It was a Tuesday night, sometime around 10 PM, sitting at Megatron with a second cup of coffee going cold next to the keyboard. I’d been working on HookHouse-Pro earlier, hit a wall, and pivoted to something I’d been putting off: setting up the WordPress Auto Writer plugin I’d been testing for Knuckledust Chronicles.
The concept is simple enough. You feed it a knowledge document about yourself, define the voice, give it rules about what not to fabricate, and it generates blog posts on topics you specify. I’d spent a couple of hours building out the reference document, the author knowledge base, the tone guide. Real effort went into it. I was thorough. I listed my family, my stack, my projects, my sports teams, the fact that Flash passed away in December. All of it.
Then I ran it on a few test topics and read what came back.
The posts were, technically speaking, correct. Accurate. They referenced real projects, real details, real parts of my life. Nothing was fabricated. The plugin did what it was supposed to do.
And I sat there staring at the screen thinking: this doesn’t sound like me at all.
Not in the way where it used wrong words or got facts wrong. It sounded like someone who had read a very thorough Wikipedia article about Frank Robinson and was now writing a thoughtful summary from a polite distance. Everything was present. Nothing was alive.
I kept trying to put my finger on what was missing, which is the kind of thing my brain will absolutely refuse to let go of. That’s not a complaint. It’s just how I’m wired. Once I notice a gap between how something is supposed to work and how it actually works, I have to understand it. I don’t want to be told the answer. I want to find it.
So I started pulling the posts apart to figure out what the AI was and wasn’t able to do with the information I’d given it.
Here’s what it could do: It could write a coherent post. It could stay on voice, roughly. It could reference my homelab by name, mention Kade, bring in powerlifting or Suno naturally. It didn’t make anything up. It followed the style rules. It avoided the words I told it to avoid.
Here’s what it couldn’t do: It couldn’t write anything that I hadn’t already written first.
That sounds obvious, but it took me a while to really land on what I mean by it. Every piece of detail in that knowledge document came from me. Experiences, opinions, preferences, projects. When the AI generated a post, it was drawing from a pool I had already filled. The knowledge document was the ceiling. The AI could reach the ceiling. It just couldn’t go through it.
When I write a post myself, something different happens. I’ll start with one thought and end up somewhere I didn’t expect, because the act of writing pulls up connections my brain makes in real time. A paragraph about Exchange troubleshooting will suddenly have something true to say about patience, or about the way I approached problems as a kid who learned everything by breaking it first. That’s not a technique. That’s just what happens when a person thinks out loud on a page.
The auto-writer can’t do that. It’s not a knock on the tool; it’s a description of what the tool actually is. It’s a very good assembler of things I already know about myself.
The Ceiling Problem
The more interesting question isn’t whether the AI wrote good posts. It’s whether that matters.
And here’s where I landed, which surprised me a little: it matters for some posts and doesn’t matter at all for others.
There are posts on this blog that are purely functional. How I migrated from NGINX Proxy Manager to Caddy. How I set up Authentik for public-facing sites. Practical walkthroughs where the value is in the accuracy and the clarity, not in whether the prose has a heartbeat. For those, an auto-writer that works from a solid technical knowledge base is actually fine. More than fine. It’s efficient.
But then there’s the other kind of post. The ones where I’m working something out while I write it. The ones that start as tech posts and end up being about something else entirely. Those can’t be assembled from what I already know. They have to be discovered. And a tool that works by reassembling what I’ve already told it about myself cannot discover anything.
I’m not being weird about this. I’m not saying AI-generated content is bad or that I refuse to use it. I set the plugin up. I’m going to use it. I already have a list of topics where it’ll do the job well, and I’m genuinely glad it exists.
But I went in with a vague assumption that the knowledge document would give the AI enough context to approximate my voice. What I learned is that my voice isn’t in the facts about me. It’s in the way I connect things I wasn’t trying to connect, in the tangents that turn out to matter, in the moments where I’m writing about Docker and I end up somewhere honest about how I learned to think.
That can’t live in a reference document. Not because it’s too personal, but because I haven’t written it yet. It doesn’t exist until I write it.
There’s something else worth saying plainly. The knowledge document I built for the plugin is the most organized summary of my own life I’ve ever written. Name, family, projects, stack, history, interests. Structured, clean, complete. And reading it felt slightly strange, the way it probably feels strange to read your own resume. Accurate, but flattened. The stuff that makes a person a person doesn’t transfer well into a bulleted list.
Flash isn’t in that document as a presence. He’s in there as a data point: Miniature Dachshund, 16 years old, put down December 22, 2025. The AI treats that the same way it treats the fact that I use Caddy as a reverse proxy. Both are facts. Neither one carries any weight unless I put it there.
So that’s the honest answer. The auto-writer works. It writes clean, accurate, correctly-voiced posts in the category where I’ve given it good reference material. I’ll use it for the right jobs and I’ll keep writing the other kind myself, because those can only come from sitting down at midnight with cold coffee and figuring out what I actually think.
The tool isn’t the problem. Mistaking the ceiling for the sky is.
