Harness Engineering with Jessica Cervi

In this episode, we sit down with Dr. Jessica Cervi, a Learning Facilitator and Subject Matter Expert on AI for MIT xPRO. She'll walk us through harness engineering, an approach focused on building the environment an AI agent operates inside rather than just the instructions you hand it. Jessica breaks down how harness engineering differs from prompt and context engineering, why guardrails and evaluation are the two load-bearing pillars of the whole thing, and how using a more powerful model as a judge can catch the failures you would otherwise ship straight to your learners. We also dig into the tradeoff every builder runs into eventually: how much freedom to give an agent, and when the stakes of a task should pull you back toward tighter control. Mentioned Links: Deploying AI for Strategic ImpactFive Pillars

On today's episode, we're joined by Dr. Bruce Cameron. He's the director of the system architecture group, a senior lecturer, and the faculty director for MIT's Architecture and System Engineering Online Certificate Program.In this episode, we'll explore the world of systems thinking, what it really means to see the connections between events, why the decisions you make early in any complex system can determine everything that follows, and how organizations consistently undermine themselves by optimizing the wrong things. From NASA to Amazon to Meta, Bruce has seen systems succeed and fail at the highest levels, and he's here to share what he's learned, including some surprises along the way.Mentioned links:Architecture and Systems Engineering: Models and Methods to Manage Complex SystemsMIT Learn