Use Cases and Building Your Workflow

Generative AI is most valuable when it removes friction from starting and transforming work, while the final judgment stays with you. In this lesson you will survey the situations where AI genuinely helps, learn to spot when it does not, and build a five-step workflow that you can apply to almost any task, from an email to a project plan.

You now have a solid toolkit: structured prompts, patterns, output control, verification and an awareness of legal and ethical limits. Chapter 5 puts all of it to work in real situations. We start with the big picture and a repeatable method.

Where AI really helps

A good rule of thumb: AI shines on tasks where a decent first version is valuable and you can quickly judge its quality. Starting from a blank page is hard; editing a draft is easier. Here are common use cases.

Communication

• First draft of an email, a message to a client, an announcement or a thank-you note.
• Rewriting for a different audience: client, management, team, general public.
• Adjusting tone: firmer, warmer, more concise, more formal.
• Preparing talking points for a meeting or a difficult conversation (with anonymized details).

Organization and planning

• Turning a pile of notes into a prioritized task list.
• Building a weekly schedule, a project outline, a checklist or a packing list.
• Preparing an agenda and, afterward, a recap with action items.
• Brainstorming options: names, themes, ways to solve a problem.

Understanding and learning

• Explaining a concept at different levels ("as if I were fifteen," then "for a professional").
• Creating practice questions from your own notes.
• Comparing two ideas or approaches.

Creating and presenting

• Presentation outlines, speaker notes, interview questions.
• Job descriptions, FAQs, templates, standard procedures.
• Ideas for visuals, with an image tool for illustrations and mock-ups.

When not to use it, or not alone

Knowing where not to use AI is as important as knowing where to use it.

• High-impact decisions about people (hiring, grading, discipline, medical or legal matters): these need human judgment, accountability and, often, regulatory compliance.
• Content with sensitive or confidential data in a tool that is not approved for it.
• Situations where accuracy is vital and cannot be verified: do not rely on unverifiable output.
• Tasks where the process is the point: learning to write, solving a problem you need to understand, a personal message where authenticity matters.
• When the rules forbid it: school, employer, client contract or platform policy.
• When it is slower than doing it yourself: for a two-line reply you already know how to write, prompting may cost more time than it saves.

The five-step workflow

To get consistent value, follow a simple method. Notice that AI appears in the middle of the process, not at the end: you frame the problem before the tool starts, and you decide what is good enough after it finishes.

1. Define: what is the exact goal, who is the reader, and what does "done" look like? How much does an error cost?
2. Prepare: gather the useful material and context, anonymize, decide what you will and will not share, and check that the tool is allowed for this task.
3. Ask: write a structured prompt (role, context, task, format, constraints), use a pattern if relevant, and iterate with precise feedback.
4. Verify: proofread, check facts and figures against sources, look for bias, and make sure nothing confidential slipped in.
5. Finalize: adapt the result with your own voice and expertise, decide whether to mention AI, then own the result and send or publish it.

The workflow in action

Imagine you must write a short report on a customer survey. Here is how the five steps could look.

1. Define: a one-page summary for the management team; they need three findings and two recommendations; an error in a figure would be embarrassing (medium stakes).
2. Prepare: export the survey results, remove names and contact details, and keep only the answers and the question wording.
3. Ask: first prompt to extract and group the main themes; second prompt to draft the summary from those themes with a fixed template; follow-up to shorten and adjust the tone.
4. Verify: recompute every percentage in a spreadsheet, and compare each finding with the raw answers.
5. Finalize: rewrite the recommendations in your own words, add context only you know, and share with a note on how the analysis was done.

Measuring whether it is worth it

Do not assume that AI saves time. Measure, at least roughly:

• Time: how long did the task take with the AI, including verification, compared with doing it alone?
• Quality: is the result better, equal or worse? Did it need less correction after a few uses?
• Risk: did you have to handle sensitive material, and were you comfortable with it?
• Effort: is it less tiring, or just differently tiring?

The honest answer might be "it saves time on first drafts but not on final checks." That is useful knowledge: it shows you where to invest in better prompts.

Capitalize: your prompt library

When a prompt works well, save it with the variable parts replaced by placeholders. A library of tested prompts is the heart of your personal work assistant. Here is an example entry:

Name: Proofread and polish
Purpose: Fix mistakes and improve clarity without changing my voice.

Role: You are a demanding but kind proofreader.
Context: The text below is intended for [audience] and will be
sent as [email / report / post].
Task: Correct spelling, grammar and punctuation. Then suggest at
most three clarity improvements.
Format: First the corrected text, then a numbered list of the
suggestions, each with a one-line reason.
Constraints: Keep my style and meaning. Add no new information.
If a sentence is ambiguous, ask me rather than guessing.

Text: [paste here]

Checks: Read the corrected text against the original for changes
of meaning. Never paste confidential data.

Your turn: apply the workflow

Exercise 1. Choose a real task from the coming week (a message, a plan, a summary). Write the five steps for it in a few lines: what you define, what you prepare, which prompt you will use, what you will verify and how you will finalize.

Exercise 2 (project). Carry out the task. Time yourself, and compare with how long it normally takes. Add the final prompt to your library, with a "Checks" line, and note whether using AI was worthwhile for this task.

See a solution

Example plan for "write a polite reminder to a client about an overdue invoice":

• Define: a short email, firm but courteous; the goal is payment without damaging the relationship; the stakes are medium.
• Prepare: invoice number, amount and due date as placeholders; no bank details pasted into the tool; client name replaced by [client].
• Ask: structured prompt with tone constraints and a maximum of 100 words; one iteration for tone.
• Verify: check invoice number, amount, date and the payment instructions in the final email against my records.
• Finalize: add the real name and a personal touch, and send it myself.

Typical time finding: the draft is fast, but filling in and checking the details takes the majority of the time. That is normal and tells you to put the facts into the prompt template as clearly labeled placeholders.