Chapter 16 — Reps

Ten reps to turn the last chapter into a plan. They move from reading the future without getting sold, to naming the human skills you will never delegate, to lighting the fuse on your capstone. This is the final set. Do not read them — do them.

Ground rules

• Type it yourself. Write your answers in your own words, in your own reps.txt. No pasting the chapter or an AI’s reply as if it were your thinking. Your fingers learn what your eyes skim.
• Try everything in a real tool. Where a rep says “open ChatGPT / Claude / Gemini,” actually open it — browser or phone. A rep you only imagined is a rep you did not do.
• Predict before you measure. Before you run a tool or the self-assessment, write down what you expect. Then compare. The gap between prediction and result is the learning.
• Keep an honest AI-usage note. Any rep where you used AI ends with one line: what you asked, what it got wrong, and what you verified. The human owns the verdict — even on the reps.
• Keep a reps.txt. One short written reflection per rep. This is graded thinking, not busywork — and this week it becomes the raw material for your capstone.

Reps 1–3: Read the future like a pro

Rep 1 — Shipped, preview, or vapor

Find three real AI announcements from the last few months (a model, a feature, a device — search “AI announcement 2026” or skim a tech-news site). For each, decide: is it shipped (you can use it today at a known price), preview (announced/limited/waitlist), or vapor (a demo or a “coming soon” with no ship date)? Write one sentence of evidence for each verdict.

Reflection: Which was hardest to classify, and why? What tipped you off that a slick demo might be further from shipping than it looked?

Rep 2 — Category versus SKU

Take one fragile, date-stamped claim about AI — e.g. “Model X does 4K video at $0.15 a second” — and rewrite it two ways: (a) a durable category-level version (“as of mid-2026, consumer video tools make short, realistic, audio-synced clips”), and (b) the same claim with an honest date-and-hedge attached. Do it for a claim in your own field if you can.

Reflection: Why does the category-level sentence age better than the SKU-level one? Where in your job would stating a fragile number confidently get you into trouble?

Rep 3 — Multimodal in your hands

Do one genuinely multimodal task in a current assistant: show it a photo and ask a text question about it, or upload a document and ask it to pull the numbers, or use voice mode to ask about something you are looking at. Note what worked and what it got wrong.

Reflection: Predict-first: before you ran it, how good did you expect the result to be? How close were you? Where did the model quietly guess instead of admitting it could not see something clearly? End with your AI-usage note.

Reps 4–6: The durable human

Rep 4 — Rate yourself in the Future-Readiness lab

Use the Future-Readiness Self-Assessment widget below the chapter. Rate yourself honestly across the eight cross-cutting skills (prompt writing, evaluating output, fact-checking, choosing models, comparing systems, responsible use, workflow optimization, collaborating with AI). Screenshot the growth plan.

Reflection: Which two skills scored lowest? Turn the growth plan into three concrete next steps — a week, a rep, a tool to revisit — that you will actually do after this course ends. Be honest; an inflated score only fools you.

Rep 5 — Name your durable five

For your own job, write one specific place each of the five human skills lives: judgment (a decision only you should make), taste (where “correct but bland” is not good enough), relationship (a conversation that must be human), accountability (whose name is on the outcome), wisdom (a “should we even?” question in your work).

Reflection: Which of the five is most at risk of quietly getting handed to AI in your workplace? What is your rule for keeping it human?

Rep 6 — The out-of-band drill

Write your personal verification rule for the deepfake age. Given a request that moves money or secrets — an urgent email or even a video call from “your boss” — what is the second channel you would use to confirm it (call a known number, a code word, a required second approver)? Write the exact steps as if handing them to a new coworker.

Reflection: Why can’t “I recognized their voice/face” be your defense anymore? Tie your answer back to the Arup case in §16.3.

Reps 7–10: Light the capstone fuse

Rep 7 — Pick the task and write the “before”

Open code/capstone-planner.txt. Fill sections 1 and 2: choose ONE real recurring part of your job, and write your current no-AI workflow step by step, timing each step honestly.

Reflection: Why this task and not a flashier one? A good capstone task is recurring, real, and yours — not the most impressive thing you could demo. Say why yours qualifies.

Rep 8 — Choose the tool and justify

Fill section 3 of the planner: name the tool AND the model/tier you will use, and why it over the others you learned this term. Name a backup tool. Hedge and date any price (mid-2026).

Reflection: What would have to be true for your second-choice tool to become the right one? This is the “choosing models / comparing systems” skill from the whole course — show it.

Rep 9 — Baseline the numbers

Open code/impact-measure.csv and fill the before_value column for at least three metrics (time per task, a quality score against a checklist you write, and rework/errors). Leave after_value blank — you earn that in the project.

Reflection: Predict-first: how much faster do you expect AI to make this task? Write the number now, before you build. You will check it against reality in the capstone — and “faster but worse” will not count.

Rep 10 — Policy and privacy check

Fill section 7 of the planner using Appendix C. Does your task touch PII, PHI, student records, or client secrets? What must never be pasted into a public tool? Is there an employer AI policy, and are you inside it?

Reflection: If your honest answer is “this task involves confidential data,” what is your plan — redact, use a local model (Week 11), or keep a step fully human? Do not hand-wave this; it is the difference between a transformation and an incident.

Done? One Last Thing.

A one-run dry rehearsal of the capstone, so tomorrow’s build is not a blank page. Do the task once, the new way, and take notes:

1. Run your chosen tool on one real instance of the task, using the prompt you drafted in section 4 of the planner.
2. Keep an agent-log stub as you go: what you asked/delegated, what the tool produced, and — the important column — every place it was wrong or you had to intervene.
3. Verify the output the way you said you would in section 6. Write down what verification caught.
4. Write two sentences: did this actually get better, and where is the human gate that keeps it trustworthy?

If you can run the task once tonight and honestly log where the machine needed you, you are ready for the capstone. That is the whole project in miniature: AI drafts, you verify, you own the verdict — and you can prove it with a log and a number.

Up next: Project 14 — Project P14: Transform Your Work (FINAL).