The argument behind the machinery, and the engineering decisions we made building it.
Everything here is written under the same rule as the rest of this site: claims sit next
to their evidence.
Around 1908, cars were built the way most software is built today: by craftsmen who held the design in their heads and made everything fit at the bench. What Ford's reorganization of work says about software in the AI age, and why the parallels run deeper than the assembly-line metaphor suggests.
Point a model at a codebase nobody on the team wrote and it will summarize any file well. What it misses is the wiring between files, and a shallow reading reads identically to a correct one. The cure is not a stronger model. It is three checks that measure comprehension against the source.
Every shop says production-ready, because the word as used costs nothing to say. Turning it into something a buyer can check takes four artifacts they can re-run without trusting you: a behavior catalog, a gate transcript, a dual-green suite, and a stated boundary.
The number you most want before a build is the one nobody can honestly give you, because price is a function of how legible the scope is. Make the scope legible and the rest of the questions answer themselves.
Process labels and portfolios are weak predictors of how a build turns out. The variable that actually tracks outcomes is whether a team can prove its work does what it was supposed to, which is also why a redesign can quietly destroy a conversion rate.
When you change development partners, the time you lose isn't the code. It's the understanding of what the code does, which lived in the old team's heads and walked out with them. An artifact fixes that.
Verification debt, the verification gap, comprehension debt: the market has named the disease from every angle. The cure is machinery that shows its work.