Chapter 5 — Reps

Eleven reps to turn “I use ChatGPT” into “I choose the right tool.” They move from breaking the one-tool habit, to running the five-question method on real tools, to assembling a stack and guarding the privacy gate. The gym this week is your own real task list. Do not read these — run them.

Ground rules

• Type it yourself. Write your task sentences, your category calls, and your worksheet entries in your own words. Your fingers learn what your eyes skim.
• Run/try everything. Actually open the tool. A rep where you only imagined using the tool is a rep you did not do. Free tiers are fine and are the point.
• Predict before you measure. Before you open a tool, write down which category you think the task wants and what you expect the tool to do. Then compare. The gap is the learning.
• AI usage note (every rep). End each rep with one honest line: which tool you chose, why it beat the runner-up, and what you had to verify or fix by hand. You choose the tool; you own the verdict.
• Keep a reps.txt. One short written reflection per rep, in a doc you keep. This is graded thinking, not busywork.

Reps 1–3: Break the one-tool habit

Rep 1 — The “I use ChatGPT” audit

List five real tasks from your actual last week where you used (or would have used) a general assistant. For each, write the task in one plain sentence, then name the category it truly belongs to (general assistant, research, document, meeting, translation, image, video, voice, music, slides, coding).

Reflection: For how many of your five was a general chatbot actually the right category — and for how many was there a specialist you were ignoring out of habit? Name the single task where switching categories would help you most.

Rep 2 — Category before brand, ten times

Here are ten one-line tasks. For each, write only the category — no brand yet. Speed matters; aim for under five minutes total.

1.  "Summarize this 38-page vendor contract and find the auto-renewal clause."
2.  "Make a header image for the fall food-drive flyer."
3.  "Get the action items and owners out of yesterday's 70-minute Zoom."
4.  "Is it true that our county's food insecurity rose last year? With a source."
5.  "Say this two-paragraph notice in Spanish for our clients."
6.  "Turn this brief into a 6-slide board deck."
7.  "Narrate this 90-second script for the website (in a voice we may use)."
8.  "Rewrite this cranky email to sound firm but kind."
9.  "Make a 15-second clip of produce being unloaded for social."
10. "Rename 300 donor-receipt files by date automatically."

Reflection: Which two did you hesitate on, and why? Hesitation usually means the task sentence was fuzzy — rewrite one of the two until the category is obvious.

Rep 3 — Name the specialist

Take the ten categories from Rep 2 and, using Appendix B, name one real tool in each category you would actually try first. Add a one-line “why this one.”

Reflection: Which category did you know the least about before opening Appendix B? That gap is where your next quality win is hiding. End with your AI usage note.

Reps 4–6: Run the method

Rep 4 — Five questions on a tool you know

Pick a tool you already use. Run it through the five-question method on paper using code/choose-your-tool.txt: Fit, Cost, Privacy, Quality, Learning curve. Score each 1–5 for a specific task.

Reflection: Did scoring change your opinion of a tool you thought you knew? Which single criterion did you have the weakest evidence for — and how would you actually test it?

Rep 5 — Five questions on a tool you have never tried

Now pick a tool in a category you rarely touch (a meeting AI, a document tool like NotebookLM, a slide tool). Sign up for the free tier, run one real small task through it, and score it on the same worksheet.

Reflection: Compare your predicted Quality score (before you tried it) to your measured one (after). Which way was your prediction wrong, and what does that tell you about trusting demos and reviews over a real trial?

Rep 6 — The Privacy veto drill

Write down four real tasks from your work. For each, answer the Step-4 gate on the worksheet honestly: does this involve names + private details, health info, student records, client/customer data, unreleased financials, or anything under an NDA? Mark each PROCEED or STOP.

Reflection: For every STOP, name what you could safely do instead (redact first? use an approved enterprise tool? do it by hand?). Cite the relevant idea from Appendix C. Practice saying the word “no” to a convenient-but-wrong tool.

Reps 7–9: Think in stacks

Rep 7 — Map a workflow into a stack

Take one real multi-step deliverable you owe someone (a report, a flyer, a presentation, a client update). Break it into a stack of specialist tools — one lane each — with yourself as the final step. Write it as a numbered relay, like the food-bank deck in §5.8.

Reflection: Where in your stack is the human step that no tool can do? Name it explicitly. If you cannot find one, you have not looked hard enough — every honest stack ends with a person owning the verdict.

Rep 8 — Feel the specialist gap

Pick one creative task (an image or a slide layout). Do it twice: once with a general assistant, once with a purpose-built specialist. Save both results side by side.

Reflection: Describe the gap in concrete terms — not “the specialist was better” but how: legibility, layout, editability, rights, time. This felt difference is why category-before-brand matters. End with your AI usage note.

Rep 9 — Cost, honestly

Take the specialist from Rep 8 (or another paid-ish tool). Estimate: how many times a month would you use it, how much time does it save each time, and what is the plan’s real cost as a mid-2026 snapshot (check the vendor page — do not guess)? Decide free tier vs paid vs skip.

Reflection: State your decision as a sentence a boss would accept: “This tool is/ isn’t worth $X/month because it saves N hours on tasks I do M times a month.” Cost is a ratio to value, never a raw number.

Reps 10–11: Run the lab and hold the verdict

Rep 10 — Score yourself on the Task → Tool Matcher

Use the Task → Tool Matcher widget embedded below the chapter. Feed it five tasks from your own real inbox. For each, predict the category first, then see what the Matcher says.

Reflection: How many did you predict correctly? Log every “there’s a specialist for this?” surprise — those surprises are your highest-value takeaways this week.

Rep 11 — Catch the confident fabrication

Feed a small factual brief (five real numbers you know are true) to a specialist tool — a slide tool or a chatbot making a summary. Then check its output against your source, line by line. Hunt specifically for: a number that drifted, a statistic that was never in your brief, a real figure rounded into a vaguer “nicer” one.

Reflection: Did it invent or drift anything? Quote the exact line. Whether it did or didn’t this time, write one sentence on why polished output demands more verification, not less. End with your AI usage note.

Done? One Last Thing.

A miniature of Project 5, end to end, in thirty minutes. Take the shipped brief code/deck-brief.txt. Do three things:

1. Choose — using code/choose-your-tool.txt, pick a slide tool and write one line on why it won.
2. Generate — feed the brief to that tool and get a first deck. Do not fix anything yet.
3. Verdict — open the deck next to the brief and list, in a short doc, every place the deck drifted from the truth: a changed number, an invented claim, a buried soft spot, a missing ask. Write your name at the bottom as the accountable human.

If you can do that in thirty minutes tonight, you can do Project 5. That is the whole job in miniature: the tool drafts, you test it against the source, you hold fast to what is good, and you sign.

Up next: Project 5 — Project P5: Deck in an Hour.