Notes from the actual work.
What I learn building AI systems for real businesses. No theory, no hype.
Why Your Team Ignores The Process You Wrote Down
The same instruction sat in four separate files, was read, was agreed with, and all 25 pages broke it anyway. Where a rule lives decides whether it holds.
Why Your AI Context File Is Probably Wrong
It never invented anything. It wrote down what I said and made it look checked, and I carried the file for weeks before a second AI session caught it.
How Do You Know If AI Actually Did The Work?
Five scripts reported success in one session and none of them had worked. Two questions separate work that happened from work that was reported.
How Do You Tell If AI Research Is Any Good?
Fifteen sources agreed with me, and all fifteen had read the same blog post. Two questions that catch this, plus the one change that turned 15 sources into 49.
How Do You Verify AI Output Before You Act On It?
A second AI agreeing with the first is not verification. The three-layer pattern I used on a court deliverable: 602 field comparisons, 6 disagreements, 0 errors shipped.
How to Make AI Remember Your Business Between Sessions
Three plain text files, no plugin and no subscription. Plus the reason a correct note still gets ignored, which I found across 11 client projects.
Want a system built for your business?
Everything here comes out of real client work. If you want something custom — a content workflow, an automation, an AI setup — let's talk.
Get in touch