How to use this book

Treat this as an operating manual, not a book to finish and forget. Each chapter answers one decision: what to delegate, which model to choose, how much effort to spend, what evidence to provide, and where a human must approve.

The examples assume real work: messy files, incomplete context, deadlines, and review obligations. Start with a reversible task. A useful first result is not the most impressive artifact; it is one whose quality you can judge.

1. Choose one recurring job with a clear owner.
2. Define the finished artifact and acceptance tests.
3. Provide a small, authoritative evidence pack.
4. Run, inspect, revise, and record what worked.

Workshop: How to use this book

The practical objective of this chapter is turning reading into a controlled learning cycle. Before opening a chat or Work, write down how the job is performed today, who owns it, and what counts as an acceptable result. Choose one example you can personally inspect. This baseline prevents the novelty of the tool from being mistaken for real improvement and gives you a fair comparison for elapsed time, accuracy, corrections, and usefulness.

Worked example: an operations lead chooses a weekly status summary as the first exercise because the source data and expected format are already understood. The team first narrows the outcome and assembles this pack: the previous human-made summary, its source files, and the manager's correction notes. It then runs the agent with an explicit stopping condition, checks every material claim, and records corrections. The example succeeds only when a named owner can explain why the result was accepted, which parts remained human work, and exactly what should change on the next run.

The most common misapplication is trying several impressive projects at once and learning nothing reliable from any of them. The correction is not automatically a stronger model. First reduce scope, rank sources, reveal hidden constraints, and add an approval checkpoint. If the problem remains after those changes, test a higher tier or effort setting on the same sample. This allows you to compare cause and effect instead of merely comparing two different-looking outputs.

• Describe the current human process in five sentences.
• Prepare an evidence pack that includes the previous human-made summary, its source files, and the manager's correction notes.
• Set one measurable acceptance rule and one prohibited action.
• Ask the reviewer to classify errors by severity, not only style.
• Save the brief, settings, output, and corrections as one run record.