Goal Engineering: Designing Agent Scopes for Better Human Feedback

LangChain argues that when designing goal-oriented agents, the scope should be chosen so that humans can easily give high-quality feedback, since unwritten expert context is the hardest thing to supply. Its case study of Actively AI shows the sales agent builder replaced per-account agents with "motions" — agents built on Deep Agents that each own a job like outbounding across many accounts — making review and cross-task learning easier.