A helpful build session, read as a warning.
This version uses the same kind of card app, but changes the argument. It treats the AI-assisted build as a case study in false confidence, drift, hidden supervision work, and the human judgement needed to stop plausible output from damaging the real goal.
This version is the critical reading of the same source material. It keeps the card grid, sequence mode, story view, and human/AI dialogue format, but it changes the claim being made. The build is no longer presented as a clean lesson in collaboration. It is presented as evidence that AI work can create plausible output while moving the burden of direction, taste, correction, and safety back onto the human.
The cards focus on ordinary failure modes: the brief gets flattened, confidence arrives before clarity, the interface makes dependence feel friendly, rhetoric can be flipped around the same facts, and a local improvement can regress the thing that made the original product valuable. The point is not that AI never helps. The point is that its help can hide supervision work and make weak output look finished.
The app is still a static single-page prototype. It has no dependencies, no build step, and no backend. Open web-app/index.html in a browser to review it.