Lecture 5: AI Taking Action
Lecture 3 closed with a warning: prompt injection gets much worse once AI starts acting instead of just writing. Here in Lecture 5, we explore that warning in detail. We open with connectors — the links that let AI read your email, calendar, or files, and act on your behalf — and show that nothing new is happening under the hood: it’s the same next-token prediction from Lecture 1, but occasionally generating a specially structured text that triggers actions. From there we build the vocabulary for real autonomous work: the analyze–act–observe loop every agent runs on, the seven-part “anatomy” of a well-specified agent, and professional-grade agentic workflow techniques like the “elephant and the goldfish.” We close by putting on a pessimistic security hat: for any agent you design, what’s the worst that could happen — and whose permissions made it possible?