Chapter 15 — Reps

Ten reps to turn a chapter of ideas into a workflow you actually run: map your week, triage every task, find what is safe to automate, design one workflow end to end, measure it honestly, and analyze a real workplace scenario. This is a drill week — the reps are the training for the graded lab at the bottom of this page. Do not read these. Do them, in the real tools, on your real work.

Ground rules

• Write it in your own words. No pasting the chapter’s tables or a chatbot’s paragraphs into your answers. The point is your week, your judgment. A workflow you did not think through is a workflow you cannot own.
• Actually do it, don’t just plan it. Open the tools. Fill the template. Run the task both ways. A rep you only imagined is a rep you did not do.
• Predict before you measure. Before you time a task or run a workflow, write down what you expect — how long, how many revisions, whether AI helps. Then compare. The gap is the learning.
• AI usage note, every rep. End each rep that used AI with one honest line: what you asked, what it got wrong, and what you verified. You own the verdict.
• Keep a workflow-notes doc. One short written reflection per rep. This is graded thinking, not busywork — and it becomes the raw material for the lab.

Everything you write into today feeds one file: code/weekly-workflow-template.txt. Copy it into your own doc now.

Reps 1–3: Map the week and draw the lines

Rep 1 — Map your real week

Fill Part 1 of code/weekly-workflow-template.txt by hand. List every recurring task, its frequency, its honest time cost (include formatting and re-reading), and its shape — rule-shaped or judgment-shaped. Do this before you ask any AI to help; look at your actual week, not a generic one.

Reflection: Which single task costs the most time for the least judgment? That is your prime automation suspect — name it and say why.

Rep 2 — Triage every task

Fill Part 2. Sort each mapped task into SELF (do it yourself), ASSIST (delegate to AI, then verify), or NEVER (never delegate / never paste). For each, give a one-line reason. Use the test: “If the AI got this subtly wrong, would I catch it before it did harm?”

Reflection: Which task was hardest to place, and why? Name one task you were tempted to put in ASSIST that actually belongs in SELF or NEVER.

Rep 3 — Draw the never-delegate line

Go back through your tasks and mark every one that touches confidential information, PII, PHI, or a decision that is legally or professionally yours to make. Cross-check against your own employer’s policy and Appendix C. Write the explicit rule you will follow: what never leaves your organization’s walls into a public tool.

Reflection: Was there a moment you realized a “convenient” AI shortcut would have crossed the line? Describe it. If not, invent the most tempting one for your job and pre-decide your answer now.

Reps 4–6: Automate, choose, and compose

Rep 4 — Run the four-part automation test

Fill Part 3. Take only your ASSIST-zone tasks and score each: Recurring? Rule-shaped? Verifiable (cheap to check)? Bounded stakes? Mark the verdict: reusable template, agent task with a human gate, or leave as a one-off.

Reflection: Which task made you type nearly the same prompt three times? That repetition is the signal — describe the template you would capture, including the fill-in blanks.

Rep 5 — Choose a model per step

Pick your strongest automation candidate and fill the model column of Part 4. Assign a model tier to each step — cheap/fast for simple or bulk, balanced for everyday, frontier/long-context for genuinely hard reasoning. Then actually run one step on a cheap tier and once on a stronger tier and compare the output.

Reflection: For your task, did the stronger tier actually produce a better result, or was the cheap tier fine? What does that tell you about where to spend capability (and money)? Flag any price as a mid-2026 snapshot.

Rep 6 — Compose a tool relay and place the gates

Design one workflow as a sequence in Part 4: mark each step AI-step or HUMAN-VERIFY, name the tool, and name what gets checked at each gate. Then build the same task in the Workflow Mapper widget below the chapter and watch whether it flags any two-AI-steps-in-a-row with no gate between them.

Reflection: Where is the most dangerous seam in your workflow — the hand-off where a dropped name, transposed number, or upgraded “we decided” would slip through? Quote the verify step that guards it.

Reps 7–8: Measure and guard

Rep 7 — Baseline and measure one task, honestly

Fill Part 5. Pick one task. Time the manual version first (predict it, then measure). Then time the AI-enhanced version including verification. Track a quality signal too: revisions to final, and errors caught later. Compute honest time saved = manual − AI-enhanced-with-verify.

Reflection: Was the honest number bigger or smaller than the hype in your head? Did quality go up, down, or hold? State your verdict — keep, drop, or redesign — and why.

Rep 8 — Guard the skill: do one cold rep

Fill Part 6. Pick one task you have been letting AI draft, and do it cold — from a blank page, no AI — once. Notice how it felt: rusty? fine? Then decide how often you will keep a manual rep to stay sharp enough to catch the machine.

Reflection: Did the cold rep reveal any skill that has started to fade? Why does keeping this muscle matter specifically for your ability to verify AI output (not just for nostalgia)?

Reps 9–10: Analyze and simulate

Rep 9 — Analyze the workplace scenario

Open code/workplace-scenario.txt — Maria’s week at Cedar Ridge Insurance. Work the analysis prompts at the bottom of that file (map, triage, automate, models, compose, measure, guard, fix-the-mess) in your own words, the way the chapter analyzed it in 15.8. Get the Protected Client Information line exactly right — that is the sharpest test in the scenario.

Reflection: Where did you disagree with the chapter’s Section-15.8 read, or push it further? Defend one triage call that a careless reader would get wrong.

Rep 10 — Run the Workflow Mapper twice

Use the Workflow Mapper widget below the chapter. Build one low-stakes task (a brainstorm) and one high-stakes task (anything with money, health, or a legal sign-off). Predict how many human-verify gates each needs before you build them.

Reflection: How many more gates did the high-stakes task need? What does that difference teach you about matching your verification effort to the stakes?

Done? One Last Thing.

A miniature of the lab, end to end. Pick one real recurring task from your week. In a single page of your workflow-notes, produce:

1. The map — the task, its frequency, time, and shape (one line).
2. The design — its steps, each tagged AI-step or human-verify, with a model tier per AI-step and a gate before anything ships.
3. The measure — a manual baseline vs an AI-enhanced-with-verify number, and one quality signal.
4. The verdict — keep / drop / redesign, in your own words, with your name on it as the accountable professional.

If you can do this for one task tonight, you can do the graded lab for a whole week. That is the job in miniature: map, design, verify, measure, sign.

Graded Lab — Your AI Week (drill lab)

This is the graded assignment for Week 15. There is no separate project file this week; this lab is it, and it becomes your Canvas submission. It merges the two halves of the chapter — design an AI-enhanced weekly workflow for your own work and analyze the provided workplace scenario — into one deliverable, plus one measured result.

Due: End of Week 15. Submit: A link to a shared doc or folder (Google Drive / OneDrive / a PDF) — not a code repo. Include your completed template, your scenario memo, and your measurement. Use placeholders for any real client names, account numbers, or confidential details; never paste protected information into an AI tool or into your submission. Allowed tools: Any consumer AI assistants and specialized tools you have (ChatGPT, Claude, Gemini, meeting/transcription tools, etc.), reachable from a browser or phone. No coding required. Keep the never-paste rules from Appendix C in front of you. AI policy: You may use AI throughout — but the design decisions and judgment calls are yours, in your own words, and every AI-touched part carries an honest AI usage note. You own the verdict.

Deliverables

1. my-ai-week.[docx/pdf] — Your completed code/weekly-workflow-template.txt, all six parts filled for your real work: the map, the triage (with the never-delegate line clearly drawn), the automation opportunities, one workflow designed end to end (steps, model tier per step, tools, verification gates), a measured before/after on at least one task, and your skill-guard habit.

2. scenario-memo.[docx/pdf] — A one-to-two-page memo analyzing code/workplace-scenario.txt (Maria’s week), addressed to “Maria’s manager.” Cover map, triage, the strongest automation opportunity, model choices, one composed tool-relay with gates, what she should measure to answer her manager’s question, one skill-guard habit, and a fix (named to the pitfall in 15.10) for each of the three complications in the scenario.

3. measurement.[any] — Your honest before/after for one real task: manual baseline vs AI-enhanced-with-verification, plus a quality signal (revisions, errors-caught-later). One line stating your verdict.

4. AI usage note — One short honest paragraph across the whole lab: what you delegated to AI, where it was wrong, and what you verified yourself.

Normal-tier rubric (out of 100)

What mastery looks like: I can read your my-ai-week and your scenario-memo and tell that you are in the loop — that you drew the confidential-data line without being told, that your gates sit exactly where a mistake would otherwise ship, that your measurement counts verification time honestly, and that you can name what you will keep doing by hand to stay sharp. The map is table stakes. The judgment — where AI belongs, where it never does, and where the verdict stays yours — is the grade.

A theological footnote. “So teach us to number our days that we may get a heart of wisdom” (Psalm 90:12, ESV). This lab is a small, concrete act of numbering your days: you are deciding, in writing, what your finite working hours are for — which of them the machine may carry, and which are reserved for the judgment, the neighbor, and the accountable call that only a person can make. The goal was never to do more. It was to spend the hours you were given on what deserves them. Design the workflow so the machine handles the counting and you are left free for the wisdom.

Up next: This is the drill week — the graded lab is above (Your AI Week), and it is your submission. When it is done, on to Chapter 16 — The Future of Work: Capstone.