Skip to content
AI Adoption Trends
Lesson 8 of 845 minutes

Build Your First AI Playbook and Know When to Use It

Bring the skills together in one practical, reviewable piece of work.

By the end: Finish a one-page AI use plan that another professional could try safely.

You have practiced a first conversation, a clearer prompt, revision, documents, research, numbers, and a small workflow. The last step is not a test of trivia. It is choosing one real job you can improve without pretending that a tool will run the business for you.

Your playbook should be short enough to use on a busy day. It names a task, the facts the assistant may use, what a good result looks like, how a person checks it, and when to stop. That is what competence looks like here: useful work, clear judgment, and a repeatable way to learn.

Your first AI playbook

The job

One repeated task and its owner

The evidence

Approved inputs and current baseline

The check

Reviewer, test cases, and quality bar

The decision

Use, change, or stop—and why

Competence is a reviewable way of working, not a collection of clever prompts.
01

Pick a job you can observe

Choose work that recurs at least weekly and produces something reviewable: a first draft, a meeting brief, a comparison table, or an internal explanation. Do not start with “use AI across the department.” The smaller task lets you compare before and after. Name a person who already owns the final result.

Write the baseline in plain language. How long does the task take now? What errors matter? Who signs off? If you have no measurement, sample five recent items. A rough, honestly labeled baseline is better than a made-up return on investment. If you cannot find five comparable items, narrow the task.

Try this

Copy a playbook-building prompt

Use this to draft a plan, then edit it with the real owner.

Help me draft a one-page AI use playbook for this task: [NAME ONE REPEATED TASK]. Ask me up to five questions first if the owner, approved information, accepted output, or risk is unclear. Then produce: (1) task and current baseline, (2) approved inputs and exclusions, (3) a reusable prompt, (4) a human review checklist, (5) five normal and three failure test cases, (6) measures for time and quality, and (7) a written pause rule. Mark anything I have not supplied as “to confirm.” Do not invent policies, savings, or test results.

Why it works: The assistant can draft the structure. The owner must supply the missing facts and approve the process.

02

Run one honest comparison

Use a small set of similar items. Have a person perform the old process and an AI-assisted process, then compare the same accepted result. Count all human minutes, including prompting, checking, and corrections. Note quality failures separately from style preferences. Do not claim cash savings because a task takes fewer minutes; that requires an actual budget action.

A good capstone may conclude “not yet.” Perhaps the source material is too messy, the reviewer spends more time checking than drafting saved, or the work is too sensitive for the available tool. That is a useful finding. It tells the team what to repair or avoid.

03

Choose a next habit, not a grand promise

If the trial works, decide who owns the prompt and source material, how often they are checked, and what changes would trigger a new test. A model update, new policy, or different customer group can change the result. Keep a short record of accepted work and corrections. Competence is maintained, not achieved once.

If you want to go deeper, the site’s workflow-evaluation path covers baselines, failure tests, and funding decisions for teams. The ROI decision lab helps you model one workflow while keeping released time separate from cash. Neither substitutes for the small comparison you can run yourself this week.

Practice assignment

Your first AI playbook

Use one real, approved low-risk task or the fictional attendee-support task from this path.

  1. 1Write the job, owner, baseline, approved inputs, and a reusable prompt.
  2. 2Test five normal items and three difficult ones; record review time and corrections.
  3. 3Decide whether to use, change, or stop the workflow and say why.
  4. 4Share the one-page plan with the person who owns the final result.

What to save

A one-page playbook plus an eight-item test log and decision.

You are ready when

The owner can tell what the AI may do, what a person must check, what evidence supports the result, and when to pause. This is a learning-path completion standard, not an accredited certification.

Further reading

Sources and guidance behind this lesson, not required reading to finish it.