In this discussion, Andi Gutmans (Google) and Peter O'Connor (Stack Overflow) explore the shift from human-centric workflows to agent-scale operations. Gutmans argues that the primary technical challenge has moved from model capability to system optimization, specifically focusing on finding the minimum context required to achieve reliable outcomes at the lowest cost. He advocates for "hill climbing" where models and data platforms are optimized together, rather than relying on static, human-curated ontologies. A critical concern raised is the "cost spiral" inherent in agentic scale; unlike humans, agents operate 24/7 and can execute tasks at much higher velocities, necessitating robust cost governance and observability. To manage this, Gutmans suggests treating the "agent" as a distinct persona within platform engineering, requiring specialized infrastructure for security, governance, availability, and cost management. This approach mirrors the evolution of data systems like Spanner, where ownership of complexity shifts from the developer to the platform. Ultimately, the goal is to move toward "intent-driven outcomes," where the complexity of data engineering and model orchestration is abstracted, allowing both developers and business users to achieve results through agentic fluency.