AI Agents and Workflow Automation

A chatbot that gives a bad answer wastes a minute. An agent that takes a bad action issues the refund, wipes the database, or sends the email — software that can *act* is held to a different standard than software that only talks. This course is about closing that gap: turning agent demos into automations you can leave running against real business systems. You'll begin where every serious automation does — mapping the actual process and deciding, honestly, what to automate and what to leave alone. Then you'll place each job correctly on the spectrum from simple script to full agent, compose the handful of workflow patterns that cover most real cases, and connect agents to tools and data through clean, least-privilege, idempotent interfaces. You'll put people in the loop exactly where stakes and reversibility demand it, and monitor everything with traces, evals, cost caps, and SLOs — alongside a clear-eyed look at prompt injection and the lethal trifecta. The through-line is accountability. When an agent acts, a human is still answerable — and the well-documented disasters, from Air Canada's chatbot to a wiped production database, are treated as lessons rather than horror stories. No coding is required, though you'll read the occasional real artifact: a tool schema, a permission gate, a trace. Across 37 lessons you'll come away with an operator's playbook built on the parts that last, not the framework of the month.