AI agents are going multiplayer. The frontier is shifting from a single assistant answering prompts to teams of specialized agents that plan, critique, and build together - which raises a stranger question: what happens when the thing agents produce is knowledge, for humans they'll never meet? I explored this by building /codelab, a multi-agent pipeline that reads a real codebase and generates interactive onboarding codelabs for the humans joining it. But a codelab is useless if it takes 10 minutes to compile, so I also built a mini Android sandbox app engineered for fast build loops at Slack, which helped a couple interns onboard during their summer to get familiar to Kotlin. This reinvents what onboarding means and how we can all align on a team while simultaneously teaching new engineers how the team works even in an agentic world.