Engineering use case: onboarding to a codebase

The strongest professional application he names (10:04):

  • Why written documentation fails at onboarding:
    • it's painful to produce and must be kept up to date (10:04);
    • it's almost certainly outside the reader's zone of proximal development — it can't be right for everyone at once (10:04).
    • Concrete mismatch cases: someone who knows your stack but not your problem domain, versus someone who knows the domain but has never used TypeScript (10:04).
  • The teach shape instead: give the new hire their own workspace, point it at the codebase, and let them learn independently how to work with the code and what the concepts are — "a productive employee in record time" (10:30).

Personal use cases

  • Side projects and learning things for fun — starting with the Rubik's Cube (10:30).
  • Chess openings, from old chess material (10:52).
  • Vocal harmonies — a skill he's always wanted but never been good at (12:36).
  • Coding languages or human languages (12:36).

The closing argument: developers are the first movers

  • AI is currently better at writing code than at almost anything else — a few exceptions aside (11:07).
  • Therefore developers hold an accidental advantage: they are the first group to test AI on a problem it is genuinely good at, and so the first to really get to know AI and build things with it (11:07).
  • The move that follows: take the ideas developed in coding, turn them into skills, and bring them out to the world — teach is his first phase of exactly that (11:44).
  • The morale note for frustrating sessions with Claude Code or Codex: the instincts you're building working with AI now transfer outside coding domains, and that will be valuable whatever the future of work looks like (11:44, 12:14).

Where to follow the skills

  • AI Hero → /skills, with a newsletter announcing new skills and changes — he notes he changes the skills all the time, so pinned knowledge goes stale (12:14, 12:36).