**Lehrwerkstatt** (German: *teaching workshop*) — a learning environment where the shop floor itself is the classroom. You learn by proximity to the work, not through a curriculum. Tobi Lütke learned programming at 16 as a German apprentice by watching experienced developers, making them coffee, and hanging around until their judgment seeped into his. Shopify's River AI agent recreates this at scale across a 10,000-person company. **Core mechanism:** Osmosis learning — knowledge transfer without explicit instruction, through sustained proximity to visible work. **Requirements for osmosis learning to function:** 1. Work must be visible to the maximum possible extent 2. Learners have access to the full stream, not curated highlights 3. No curriculum or training plan required — the environment teaches **Why this scales with AI:** - Junior developers watch how senior engineers scope requests before they send their first one - Support engineers learn log query patterns by watching backend engineers in adjacent channels - Each team's accumulated taste flows into the agent; the agent gets better at being Shopify - Merge rate improvement (36% → 77%) came from observation and skill-writing, not model changes **Contrast with private tool use:** When AI interactions happen in private windows, only the person at the keyboard learns. The apprenticeship is locked to one person. Knowledge stays siloed. **Applications:** - Education design: open learning environments over private tutoring wherever possible - Engineering teams: public code review and pair programming as core learning infrastructure - AI deployment: treat session visibility as a first-class design constraint