A notebook, in public
The side with no pitch on it.
I test AI models on a Tuesday and write about them on a Wednesday. Most of what I know I learned by running things, first inside big companies for ten years, then in a company of my own. I keep the writing here in plain words, with the numbers I saw, and I say when I was wrong.
Photographs
Writing
Now
- →Learning AI by running it. I put every major model release onto real work the day it lands, and log what got faster, what broke, and what it cost.
- →Solving growth for three kinds of company: a $48M consumer service business where I run AI and growth, B2B founders at $5M to $100M through NuroSparx, and a SaaS product I'm building on the side.
- →Getting brands cited by ChatGPT, Gemini and Google AI Overviews, then measuring whether that traffic converts. So far it does, at about five times Google's rate.
- →Sharing all of it in public: 30 X articles in 30 days (14 done), The Other Brain once a month, and ten comments a day on other people's posts.
- →Running five brands on 12 research units a month with a small team, and teaching the team to run the system without me.
Where I was wrong
- →Aug 2026. I thought faceless accounts could carry a brand. They got reach and lost trust. I put my face back on.
- →Jul 2026. I thought accurate AI-written content would rank. AI never cited it, and it never ranked. AI models cite original numbers.
- →Jun 2026. I skipped posting for 19 days and assumed the audience would wait. The month's growth went to zero.
Predictions I'll be graded on
- →By Sep 2027: fewer than 300 of every 1,000 US Google searches end in a click to the open web. Down from 360 in early 2026.
- →By Mar 2027: a $5M to $100M company that ChatGPT and Gemini don't cite for its category loses a measurable share of inbound. I'll publish the numbers from the brands I run.
- →By Dec 2026: most agent projects at mid-market companies still die before month five. Median time-to-value stays above four months.