Draft: Neither the schedule nor the content has been decided. The sessions may change.
Consider where value stalls, and who controls what, as you hand more work to AI. Designed as training for working professionals, it runs over three days (twelve 90-minute sessions). You will study what autonomous AI can do and its risks, estimate costs and results, find where adoption gets stuck, and design controls through identity, permissions, approvals and records. Oversight is designed not by approving every action, but in four fields: what is delegated, when to stop, what people check, and how records are kept. Finally, you will summarize your organization's AI adoption plan, including its controls, on one page. For students, a smaller six-session version is offered.
01Part 1: Capabilities and risks (Day 1)
Session 1What autonomous AI can do, and what can go wrong
Session 2Reading official warnings and classifying the risks
Session 3Reading costs from how AI works: tokens, inference and model size
Session 4Cost per success: a three-layer table of costs and results
02Part 2: The adoption process and designing controls (Day 2)
Session 5What counts as "in production": reading adoption statistics
Session 6Connect, context, control, adoption: predicting where your organization gets stuck
Session 7Designing identity, permissions, and checks on input and output
Session 8Designing human approval and tamper-proof records
03Part 3: Hard problems and organizing for them (Day 3)
Session 9Tabletop exercise: poisoned memory and a monitor that misses things
Session 10Measuring whether approvals have become a formality
Session 11Reading benchmark numbers and vendor changes, and deciding what to keep in-house
Session 12Putting the adoption plan on one page and presenting it to leadership