Ever since I can remember, I've wanted to be an AI engineer. after Goodfellas.
SEVEN DAYS. ONE NEW VERTICAL. PRODUCTION SHIP.
EIGHT PHASES. DOCUMENTED. REPRODUCIBLE.
NOT AN AGENCY. NOT A FRAMEWORK. A FACTORY.
The factory is the answer to a question I lost sleep over: how do you ship a new client vertical without rebuilding the platform every time. Once the platform is separate from the vertical, the repeat work stays repeatable. What changes per industry is the language, the compliance surface, and the escalation rules. What shouldn't change is the platform.
Seven days is the target when the platform holds. Day one is discovery — the industry's actual working shape, not the sales-deck version. Two through five is the vertical pack: prompts, gates, escalation rules, compliance rewrites. Day six is canary. Day seven is production — or a written reason it isn't ready.
Eight phases means the process is written down. Written down means it can be inspected, challenged, and run again without pretending every vertical is a blank page. If a phase drifts between verticals without a reason, that's a defect in the factory.
Not an agency. Not a framework. A factory: reuse what should repeat. Rebuild what the vertical proves is different. Show the work.
The factory didn't start at seven days. Early verticals missed it — I kept touching platform code when the work was really vertical-specific. The cadence improved when I got more disciplined about what to leave alone.
Compliance is the part the factory can't pretend to absorb. It can catch the shape — opt-in, retention, escalation, approved language. It can't replace substantive review for a regulated vertical. That review is part of the clock.
Canary broke before it became useful. I ran one day-six test against traffic that looked too much like the discovery sample. Real traffic was different enough that escalation rules had to be re-scoped during launch. The lesson: a canary that only repeats discovery isn't a test.