Governing AI Agents: Visibility and Control

AI agents often look impressive in demos but struggle when they hit production, where you need visibility, safety controls, and clear accountability for every action an agent takes. In this hands-on course, Kesha Williams—a machine learning technology leader with 25+ years of experience—shows you how to work in Python to transform an ungoverned shopping agent into a governed system that behaves predictably in real-world environments. Through hands-on coding in GitHub Codespaces, learn how to add structured logging to make agent behavior observable, implement runtime guardrails that block unsafe actions, and introduce human-in-the-loop approval workflows for high-risk changes. Kesha also demonstrates how to build an agent inventory and a reusable deployment checklist that you can adapt to your own framework, giving you a practical governance tool kit—whether you are shipping your first agent feature or hardening an enterprise AI workflow.

Learning objectives
Examine the structure and execution flow of a prebuilt AI agent to understand how it reads data, calls tools, and performs actions.
Implement runtime guardrails that prevent forbidden actions, protect sensitive data, and log policy violations before an agent executes a tool call.
Modify an AI shopping agent, adding structured logging to capture actions, tool usage, and decision points for traceability and post-run review.
Use logs and state inspection to detect, trace, and analyze unintended or unsafe agent actions in a realistic failure scenario.