👋 Hey {{first_name|there}},
Six issues ago, we started pulling apart what it takes to put AI into a product that already has paying customers. This one puts the pieces back together, in a different order than I wrote them.
Why this matters
Writing order and running order aren't the same thing. I introduced these ideas in the sequence that made them easiest to explain, one concept at a time, each one setting up the next. If you run an assessment that way, you'll waste most of it.
The reason is that some of these checks are disqualifiers and some are design work. A cost model can kill an idea in twenty minutes. So can a look at where tenant data would have to travel. Evaluation strategy, failure paths, rollout planning: those are real work, and they're wasted on an idea that was never going to survive the cheap checks.
So the order is a filter. Front-load anything that can eliminate an option quickly, and don't do careful thinking about a feature until it has earned careful thinking. Most teams do this backwards. They fall for an idea, spend a month designing it properly, and discover the problem at the end, when sunk cost has already made the decision for them.
Two weeks, if you're focused. Week one decides what you're building. Week two decides how it ships.
🧭 The shift
From: "Let's analyse our best AI idea thoroughly."
To: "Let's eliminate as many ideas as possible as cheaply as possible, then analyse whatever's left."
The goal of the first week isn't a decision. It's a short list you can defend, arrived at by ruling things out rather than by falling in love.