~/VibeHandbook
$39

Chapter 19

Agent Orchestration

For most of this book the AI has been a tool you hold in your hand: you ask, it answers, you check the result. That's still the right mental model most of the time. But the tools have grown a second mode — one where you hand over a goal instead of a step, and the AI works for minutes (sometimes longer) before it comes back to you.

This chapter is about that shift, and about not losing the plot when it happens. The temptation is to read "agent" as "I can stop paying attention now." That's the reading that gets people in trouble. An agent is leverage on work you already understand — not a license to ship code you've stopped looking at.

  1. 19.1The Three Tiers
  2. 19.2When to Let It Run vs. Steer Manually
  3. 19.3Scoping a Task So It Can't Wander
  4. 19.4Background and Parallel Agents
  5. 19.5Multi-Agent: Fan-Out, Adversary, Synthesis
  6. 19.6Reviewing Output at Scale
  7. 19.7Cost and Runaway Loops
  8. 19.8Honest Limits
  9. 19.9Recap & Practice

Want it offline?

Get the PDF + EPUB + downloadable prompt library + version updates.

$ Get the PDF — $39