Choose the work you need
32. Build Your First Automation Around One Finished Job
Build a bounded automation with clear inputs, a dry-run preview, exact approvals, duplicate handling, and a runnable local example.
33. Give AI the Permission It Needs—and No More
Control AI actions with verified identity, record scope, exact proposal approvals, expiry, revocation, and checks outside the model.
34. Understand Agents, Tools, and MCP Before Connecting Your Business
Understand agents, workflows, tools, and MCP so you can choose an appropriate architecture and evaluate real business integrations.
35. Give Your AI Reliable Memory Without Teaching It False Facts
Design reliable AI context and memory with source references, access boundaries, explicit status, and separation of facts from proposals.
36. Know What Happened—and Recover When the Workflow Fails
Monitor automation quality and effort, reconcile uncertain outcomes, control retries, and practice rollback and recovery after failures.
Put this section to work
Start with the first chapter if this is new to you. Bring one real task, use permitted information, and write down the result you want before testing. Keep assumptions separate from observed results.
Use the worksheets for this section to record the work. A completed exercise is preparation for a pilot, not evidence that a live process has succeeded.