A talk by Phil Hetzel from Braintrust arguing that agentic AI development should not be siloed exclusively within data science or ML engineering teams. He contrasts traditional ML workflows (model training, feature engineering, cross-validation) with generative AI development (pre-built models, prompt/context engineering, API integration), noting that the skills required are fundamentally different. He makes the case for diverse, cross-functional teams that include product engineers, systems engineers, domain experts, and non-technical subject matter experts alongside data scientists. Data scientists still add value through guardrails, LLM-as-judge evaluation, and fine-tuning, but the broader agent-building process benefits from people with closer proximity to the problem being solved.