What changed

Why this page comes late

Seven principles precede this one, and none of them mention technology. That order is deliberate. Technology introduced before the problem is a pitch; introduced after, it is an explanation. Anyone selling AI as the story has it backwards — AI is the newest part of the shift, not the whole of it.

The real change is architectural, and it accumulated for twenty years before the loud part arrived.

The quiet accumulations

Cloud computing removed the server room, and with it the capital expenditure that made every system a committee decision. APIs turned software from sealed boxes into things that expect to talk. Composable architectures and reusable components turned building into assembly: authentication, permissions, documents, reporting — solved once, used everywhere. Continuous deployment turned change from a release event with a rollback plan into a Tuesday.

Each of these alone reads as an increment, which is why the shift was easy to miss. Together they dismantled, piece by piece, every assumption behind the old buy-or-build choice: that custom means from scratch, that change means risk, that software is something a company installs and then survives.

Then AI arrived

Onto that architecture, AI landed as a cost collapse. Code generation collapses the price of the genuinely custom layer. Configuration by description collapses the price of adapting what exists. Workflow automation collapses the price of the glue that used to be a consulting line item. The work did not get slightly cheaper — whole categories of it stopped being scarce.

The psychological consequence runs deeper than the economic one. Every business carries a backlog of ideas it never wrote down because they felt too expensive to say out loud. When execution gets cheap, that internal filter recalibrates, and ideas that died on contact start becoming experiments.

Architecture still decides

If AI writes code this cheaply, why not skip platforms entirely and generate every company a bespoke system from scratch? Because that road has been walked before, at slower speed. Generated code you own alone is custom software with faster typing: no shared foundation absorbing security, performance and infrastructure work, every improvement re-implemented per company, the freeze setting in the day the generating stops. The silo problem returns — at generation speed.

Cheap code makes architecture more important, not less. Architecture is what decides whether all that new speed compounds into a platform or accumulates into the next generation of legacy.

AI changed the economics. It did not change the physics.