Detailed teaching notes for a 3-hour course. Capability-first, theory-on-demand. The harness is drawn as a rough skeleton up front — a loop, a memory, a context window, the model — then progressively enriched as each use case demands it. Anchor thesis: people conflate the LLM with the app; the harness is ~50% of the product.
Paste this back into the chat when you converge. The [x]/[ ] marks tell me what's in and what's parked.
If this resource maps to a real bottleneck on your team, ask Tomer about a Claude Code workshop. The fit check is short: team size, current Claude usage, main bottleneck, and timing.