Translate concrete business problems such as sponsor fit assessment or dynamic pricing into agent specifications: scope, inputs, outputs, framework logic, and guardrails, Design the agent's reasoning structure and prompt/workflow architecture in close collaboration with Tech: single assistant vs. retrieval-driven vs. multi-step agent with tool access, and the orchestration logic in between, Define the knowledge slice each agent needs and work with the Knowledge Architect to ensure it exists at the required quality and granularity, Build and run evals: golden test sets, quality gates, A/B tests in production, and feedback loops from Key Account Management and end users, so iteration is structured rather than vibes-based, Own the agent backlog and the value × complexity prioritization, and lead the conceptualization work for new agents as the library grows