Small-cohort AI training program — 5 weeks, 7-10 students, active-recall + Socratic

A 5-week program, 1-2 hours per day, mini-groups of 7-10 students per cohort, one personal instructor, live lectures, homework reviews, and 1:1 instructor sessions. Curriculum: LLM fundamentals → developer concepts → hands-on Claude Code. Reported track record across the source: 14 cohorts, ~200 students, including YC founders and Google/Meta/Accenture engineers — completion + success rate near 100%.

The reason corporate AI training has ~15% retention is that it's run like marketing webinars. The reason this design has ~100% is that active recall + Socratic + small group makes free-riding impossible. The per-student cost is much higher than a webinar; the per-completion cost is far lower.

1. Cap each cohort at 7-10 students. Past 10, the 1:1 instructor ratio breaks and retention drops to webinar levels.
2. One personal instructor per cohort. Not a TA pool — one person who knows every student.
3. 5 weeks × 1-2h/day cadence. Long enough to build muscle memory; short enough that motivation doesn't fade.
4. Active recall as the dominant exercise format. Not "watch and follow along" — quiz format, "explain back to me," forced retrieval.
5. Socratic 1:1 sessions weekly. Instructor asks rather than tells; students discover the answer.
6. Curriculum sequence: LLMs how-they-work → developer concepts (CLI, git, code structure) → Claude Code hands-on. Skip steps and the non-engineers drop out at week 3.
7. Final demo day: each student pitches their own real use case.

• Don't dilute group size to maximize revenue. Multiple operators have tested larger cohorts; past 10 students the format collapses into a standard course and retention drops back to ~15%. The active-recall + Socratic format only works when the instructor knows each student well enough to call on them by name and follow their week-over-week progress.