1. What is this model designed to miss?
Successful novice-to-product paths that never look like misdirected apprenticeship: people who ship narrow tools because the intention was already specified; teams where a human partner supplies taste and architecture; cases where the product is the learning artifact and “the version I can already see” was always fantasy. It underweights distribution, users outside the builder’s head, capital, and platform policy. It misses gradual skill transfer that arrives through the game rather than only through a named human master.
2. What would disconfirm this?
Novices using only the default agent loop, with no human mentor, repeatedly reaching products whose internals they can diagnose and change — on problems that began as vague mental pictures. Or default products that force curriculum checks and still match mentored builders. Partial: someone who started unable to name the miss later gained the sight to choose the next lesson, without becoming a traditional engineer first.
3. What are 2 alternative explanations?
A. Spec and product-definition problem, not apprenticeship geometry. Unclear products fail for experts too; AI only made failure cheaper. B. Selection and phase. The author is mid-funnel on hard open-ended work; inexperience plus ambitious scope is being generalized into a theory of AI apprenticeship.
4. What is this framework optimized to make invisible?
Competence that accrues without feeling like craft education. The success class of well-bounded work, admitted only as “chores.” Collaborative paths. Economic and social outcomes of becoming someone other than the intended product’s builder.
5. What would I need to believe for the opposite conclusion to be correct?
That the default loop is already a good enough apprenticeship for product-shaped work; that “the version I can already see” is often illusory for everyone; that the runnable product-as-probe is how intention gets specified; that human masters are accelerators, not the thing that closes a gap the game cannot close.
Objections the critic marked as capable of misleading
- Single trajectory treated as medium structure — the aphorisms read as general laws; evidence is one unreleased journey.
- Unfalsifiable “right skill / right lesson” — defined from an endpoint the author cannot see.
- Product imagined in the head as the truth standard.
- False contrast: models “can” versus ordinary loop “does not.”
- “Mastery” of wrong things asserted without showing mastery.
- Human master as the closer of the gap.
- Chores versus real products binary.
- Archimedes line overpromises.
- Experts “stuck later” still recenters novice exceptionalism.
- Jet metaphor stacks opacity.
- “Limiting factor between your own ears” can absolve the tools.
The verifier later sustained 1 and 3, and did not sustain 2, 4, 5, 6, 7, or the jet line. That adjudication is on the audit page, not here.