Six years using these systems. Now I am learning how they actually work.
I have built and handed over three multi-agent pipelines, and I can tell you what each one costs per run. That turned out not to be enough. So I am going down a layer, to the maths and the machinery, and writing down what I find on the way.
From the Journal.
How I Got Into Content Marketing and Link Building
Content marketing and link building are the niches I settled into, but I didn't plan for any of it. I got in sideways, through a job I almost didn't take.
Read the piece →
I labelled the training data before I ever built with the models.
I started in sales at twenty, working physical stalls, pushing the products nobody else wanted to push. You stand in front of strangers all day, you get told no, you keep going anyway.
Then I moved online. Transcription first, then essay writing, accounts closed without warning, long stretches broke. In 2021 I spent a year labelling training data for large language models. At the time it was just work. It turned out to be the most useful year I have had, because I saw how these things are assembled and where the seams are, years before anyone was selling them to me.
Now I do SEO and content operations for a portfolio of personal brands, and I build automation systems on my own time. What I took from all of it is tenacity, and one thing I had to learn the hard way: if I wanted to make it, I had to upskill on purpose.
Read the full story →Flow Blueprints
How AI systems actually get built, run and handed over. I take one thing apart at a time: a system, a procedure, or something I got wrong. You get what it cost, where it leaked, and what I would hand to somebody else.
