Field notes from the practice.
Short, opinionated, written by the people doing the work.
Field Notes is the lab's running journal. We publish three kinds of pieces: Working Papers (long-form arguments, opinionated), Field Reports (annotated write-ups of builds we have just shipped), and Specimens (real agent outputs with our commentary attached).
Five categories are in regular rotation: The Build (engineering and architecture), The Math (build-versus-buy economics, pricing), The Craft (writing, brand, design discipline), The Forecast (where the category is heading), and The Cases (project write-ups). Roughly one piece a week. Authored by the lab. No ghostwriting and no SEO filler.
Custom GPTs are not custom AI. The difference, with diagrams.
The system prompt is the floor, not the ceiling. What you actually need for a brand-grade agent — tools, memory, evaluations, the lot.
Build versus buy for marketing AI: the math nobody puts on a slide.
What we actually charged Marcus. What it would have cost them to build it. What the SaaS shelf would have cost them at year three.
Why your brand voice keeps drifting in AI output, and the lockfile that stops it.
The drift problem is a context problem. The schema we use across every gimmefy customer to hold voice steady across a hundred thousand generations a month.
The six-week white-label: a working calendar.
What gets done in week one. What gets done in week six. The two weeks where almost everything goes wrong, and how to plan for them.
An argument against the term "AI strategy."
You do not have a search-engine strategy. You have a business strategy that uses search. The same applies here, and we should say so.