Enterprise AI is moving beyond single-prompt chatbots toward agents that plan, use tools, and complete real work—and their success depends far more on context than on the underlying model. This session shows how to architect multi-agent workflows grounded in organisational knowledge, human oversight, and well-structured SKILL.md files. The American Board of Pediatrics will share examples of setting up agents with training material and supervision, plus continuously learning systems for comment classification mapped to EPAs and bias detection and removal. The Emerging Technologies Department at the American Board of Internal Medicine will discuss a complementary approach: passively capturing context from everyday work—meeting transcripts, notes, and documentation—into a shared knowledge base that engineers and agents alike can draw on, with human review keeping it trustworthy.