I use AI to write or help write many of the articles on this site, and I’d rather tell you that in the first sentence than have you work it out in the third paragraph. What I want to explain is why, because the why is the whole point of the site.
I believe businesses should be built AI-first. Not AI-assisted, where a person does the job the old way and a chatbot tidies the edges, but AI-first, where you start from the question of what a system can do and reserve people for the parts that genuinely need them. I’ve spent thirty years telling business owners that the repeatable work belongs to systems and the thinking belongs to them, and these pages are where I write that argument down. It would be an odd place to make an exception. If I believe it, I should run it on my own work, where I can’t hide behind a client’s constraints when it gets hard.
What happens before I see a draft
The idea that an article is produced by typing a prompt into Claude or ChatGPT and pasting the answer is about as accurate as the idea that a company is produced by registering a name. The prompt is the smallest part. What runs behind the site is a workflow, and most of it has nothing to do with writing.
It starts with research. An agent reads a few hundred sources every morning, newsletters, feeds, the conversations on X, and files the stories that fit the themes I care about alongside the notes I jot down myself. On Sundays an editorial review goes through that backlog and decides which ideas deserve an article and which angle I’d take on each one. It writes a brief and lays out the week. The brief matters more than people expect, because it records the view, not the news, and the view is the article.
Then a draft is written. The writing is calibrated against my own work from the years before any of this was possible, the pieces I wrote by hand, so that the cadence is mine instead of the default register every model drifts towards. From there the draft goes through more review than anything I ever published on my own. A linter looks for the shapes readers recognise as machine-written. An editor, deliberately a different model from the writer, reads it against the brief and sends back what’s missing. A deeper pass then checks that the piece delivers what was commissioned and that nothing in it has been invented. That last rule is absolute. An anecdote or a client story appears under my name only if it comes from something I’ve actually written or lived, and if the draft needs lived experience that the record doesn’t hold, the draft waits.
Only then does it reach me.
Where I come in
I read every article cold, on my device, as a reader would, before it’s published. I choose the topics and I set the views the articles argue for. I send drafts back with notes, and I publish with one tap or I don’t, because nothing goes live without that tap. After publication another agent looks after the search side, the internal links and metadata I used to pay people to do. A further one clears the operational chores, so that my involvement stays on judgement rather than plumbing.
The output isn’t perfect. Some days I read a draft and a sentence has the faint shape of a machine about it, a rhythm a little too even or a point landed a little too neatly, and it goes back. The writing has got noticeably better over the year as the workflow has improved, and I expect that to continue, but you’re the judge of that, and I’d rather you knew what you were listening to.
Why say any of this
Partly because it’s honest, and honesty is cheaper than the alternative. Mostly because the question I’m asked most often about AI is whether it can be trusted with real work, and my answer is that it can when it’s placed inside a system with clear boundaries and a human who owns the outcome. This site is that system. Every article is a small demonstration of it, and so is this page.
The views are mine and the arguments are mine. If an article doesn’t say what I think, it doesn’t go out, and that’s the loop the human sits in.