Project P14 — Transform Your Work (FINAL)

“Commit your work to the LORD, and your plans will be established.” — Proverbs 16:3 (ESV)

Chapter: 16 — The Future of Work: Capstone Due: End of Week 16 — take-home deliverables submitted, then a live presentation in the final session. Submit: A link to a shared folder or document (Google Drive, OneDrive, or a single combined PDF portfolio) containing your capstone plan, your before/after measurements, your agent-log, your reflection, and your Hard-tier memo if attempted. Deliverables are .docx / .pdf / .xlsx / .txt documents plus screenshots — not a code repository. See Appendix A for account and submission mechanics. Allowed tools: Anything you have learned — any consumer AI assistant (ChatGPT, Claude, Gemini, Grok, Perplexity), image/voice/video tools, a local model, and agent modes (Week 13). This is open-AI and agentic on purpose. Reach everything from a browser or phone. AI policy (FINAL — Phase 2): Use AI hard. You must keep an agent-log recording what you delegated, what the tool did, where it was wrong, and where your judgment overrode it. You are graded most on the document you write first and the judgment you show — not on how much the machine did. The human owns the verdict.

The Setup

Meet Dana, a communications coordinator at a mid-size nonprofit. Every month Dana builds the donor newsletter and its companion email: pulling updates from three program managers, drafting the stories, fitting them to the brand voice, checking every name and number, and shipping it to 4,000 donors. It eats the better part of two days, it is late as often as not, and the quality wobbles with how tired Dana is that week. It is not glamorous. It is not going away. And done badly, it costs the organization real trust and real money.

That is a capstone task: real, recurring, and yours. Dana does not need a flashy AI demo. Dana needs that specific two-day slog to become a half-day of confident, verified work — with the donor names still correct, the voice still Dana’s, and a human still reading every word before 4,000 people do.

Now cast yourself. You are not a programmer and you will not write code. You are the office manager, the teacher, the nurse, the marketer, the analyst, the small-business owner — whoever you actually are — and you have one recurring part of your job that is real, repetitive, and worth transforming. This project is you doing for your own work exactly what Dana is doing for the newsletter: redesign it around AI, verify it, keep it inside policy, measure the before and after honestly, and stand up and show what you built.

This is the final. It is the whole course in one deliverable — choosing the tool, writing the prompt, checking the output, respecting the policy, keeping the human in the loop — pointed at the one thing that will pay you back every single month after this class ends.

Setup (the starter)

This chapter ships two starters in code/. They are your scaffolding; make them yours.

• code/capstone-planner.txt — the planning worksheet you fill in first, before touching a tool: the task, the “before” workflow, tool choice and justification, prompts, the redesigned workflow with its human gate, verification, policy check, and the verdict line you sign. This becomes your capstone-plan.docx.
• code/impact-measure.csv — the before/after measurement template: time, quality, rework, volume, cost, effort. Fill the before_value column now; earn the after_value column by building. This becomes your impact-measure.xlsx.

Copy each into your own document. Do not submit them blank or unchanged.

Learning Targets

You will demonstrate that you can:

• Choose deliberately — pick the right tool and model tier for a real task, and defend the choice against the alternatives you learned.
• Redesign, not just accelerate — build an AI-enhanced workflow with a clear human-verification gate, rather than paving a broken process.
• Verify relentlessly — catch and correct what the AI gets wrong, and prove you did.
• Respect the guardrails — keep confidential data out of public tools and stay inside policy (Appendix C).
• Delegate with accountability — use agents and keep an honest agent-log; own the outcome regardless of how much the machine did.
• Measure honestly — show a real before/after, where quality did not quietly fall to buy the time saved.
• Exercise judgment — decide what to transform, what to leave human, and what to recommend to others.

Normal Tier

Goal: Take one real recurring part of your job and transform it with AI — planned first, built, verified, measured, and presented.

Required features

1. capstone-plan.docx — the completed code/capstone-planner.txt: the task and why it recurs, the honest “before” workflow with times, your tool + model-tier choice with justification, your reusable prompt, the redesigned workflow with [AI] / [ME] steps and the human gate marked, your verification plan, and the signed verdict line.
2. The redesigned workflow, actually run — do the task the new way on at least one real instance. Capture the AI’s output and your final, verified version (screenshots or exported documents).
3. impact-measure.xlsx — the completed code/impact-measure.csv: at least three metrics with honest before and after values (one must be a quality measure against a checklist you wrote, so “faster” cannot hide “worse”).
4. agent-log.txt — what you asked or delegated, what the tool/agent did, where it was wrong or you intervened, and what your verification caught. This is required and it is where I look for whether you were actually in the loop.
5. policy-check (a page in your plan) — using Appendix C: what confidential data this task touches, what must never be pasted into a public tool, and how you stayed inside policy.
6. reflection.docx (400–700 words) — what got genuinely better, what you would trust the AI to do and what stays human, where the tool nearly cost you, and your three next steps from the Future-Readiness growth plan.
7. A live presentation (~5–8 minutes) — walk us through the before, the after, the numbers, and one honest place the AI was wrong and you caught it.

Normal-tier rubric (out of 100)

Medium Tier (+up to 25% extra credit)

Make the transformation deeper and more durable.

1. Two tools, one task. Run the same task through two different tools or model tiers (e.g. a flagship vs a cheaper tier, or two assistants). Compare quality, speed, and cost as a mid-2026 snapshot, and recommend which you would standardize on and why.
2. A reusable, tested prompt. Turn your prompt into a proper template with {{placeholders}} (Week 12), add it to your prompt library, and show it working across three different instances of the task — proving it is reusable, not a one-off.
3. Teach it forward. Produce a one-page “how to run this” guide a colleague with your AI literacy could follow to do the task your new way. If they could not, it is not a workflow yet — it is a trick you got lucky with once.

Hard Tier (+up to 25% additional extra credit)

This is the judgment the machine cannot make for you, and it is graded as such.

Write recommendation-memo.pdf (one page, addressed to your manager, principal, practice lead, or business partner) that makes and defends a single decision: should this transformation be adopted beyond you — by your team, department, or whole organization — and what must stay human? You must:

• State the recommendation plainly (adopt / adopt-with-limits / do-not-scale) and the evidence from your own before/after that supports it.
• Name at least one step that must never be automated on this task, and say exactly why — tie it to one of the durable five (judgment, taste, relationship, accountability, wisdom).
• Name the failure mode you are accepting, and the accountable human — by name or role — who owns the outcome when the AI is wrong.
• Tie it explicitly to the spine rule — you choose the tool; you own the verdict — and to one concrete control (a human gate, a policy line, a verification step).

A memo that says “automate everything” or “AI can’t be trusted, don’t bother” fails. The grade is in the discrimination: showing you can tell the part worth scaling from the part that must stay human, and defend the line to a decision-maker. No agent can write this for you, because it requires you to stake your own name on a call about your own workplace.

Submission

Put everything in one shared folder or a single combined PDF and submit the link (Google Drive, OneDrive, or a PDF portfolio) per Appendix A. Suggested contents:

capstone/
  capstone-plan.docx          (the planner, completed)
  impact-measure.xlsx         (before/after, ≥3 metrics)
  agent-log.txt               (required)
  reflection.docx             (400–700 words)
  before-after-screenshots/   (the AI output + your verified final)
  presentation.pdf            (your slides or notes)
  how-to-guide.pdf            (Medium tier)
  recommendation-memo.pdf     (Hard tier)

Do not submit a GitHub repository — you are a professional, not a programmer, and your deliverables are documents. Redact real names, donor data, patient details, student records, and any secret before anything leaves your machine. If a screenshot would leak confidential data, blur it or use a stand-in.

Hints (Read Before You Begin)

• Write the plan first, then build. The planner is not paperwork; it is the thinking. People who open a tool before they open the planner build fast and shallow and have nothing to present.
• Pick a boring, recurring task. The best capstone is the unglamorous thing you do every week, not the most impressive thing you could demo once. Boring-and-recurring is where the payoff compounds.
• Measure the “before” honestly, tonight. You cannot show improvement you did not baseline. If you wait until after you have optimized, your “before” number becomes a guess, and a guess is not evidence.
• Keep the agent-log as you go, not after. Reconstructing where the AI was wrong from memory is how you accidentally launder its mistakes. Log the interventions live.
• Guard the quality metric. The whole project fails quietly if you got faster by getting worse. Write your quality checklist before you build, and score against it both times.
• Name what stays human before the tool tempts you otherwise. Decide your human gate up front. It is much harder to add a gate after the speed feels good.

What Mastery Looks Like (Beyond the Rubric)

A mastered submission is one where I can read your plan, your log, and your memo and see you in the loop the whole way — that you chose the tool for reasons, that you caught the fabricated figure the fluent draft slipped past you, that your before/after is honest enough to include a metric that did not improve, and that you can name, with a straight face, the one part of this task you will never hand to a machine and defend why. The time saved is the headline. The judgment is the story. Show me the story.

Coach’s Note — The temptation on the final is to make the transformation look total — “AI now does the whole thing.” Resist it. The most impressive capstones I grade are the honest ones: AI does these four steps, brilliantly; I do these two, always, because a human has to. That is not a smaller result. That is the entire thesis of the course, demonstrated on your own work. Sixteen weeks for this. Land it.

When You’re Done (a short checklist)

• The planner is complete and the verdict line is signed with your name.
• You ran the redesigned workflow on a real instance and kept the before and after.
• impact-measure.xlsx has ≥3 honest metrics, including a quality measure.
• agent-log.txt names where the AI was wrong and where you intervened.
• Policy/privacy is documented; nothing confidential left for a public tool.
• reflection.docx says what stays human and lists three next steps.
• Your presentation includes one honest “the AI was wrong here” moment.
• (Hard) recommendation-memo.pdf defends one decision and one never-automate line.
• No real names, PII, PHI, or secrets in anything you submit.

A theological footnote. “Commit your work to the LORD, and your plans will be established” (Proverbs 16:3, ESV). The Hebrew says roll your work onto the LORD — do it well, then hand the outcome to stronger shoulders. This capstone is a small rehearsal of exactly that. You will build a workflow around tools that will have changed before you present it, for a future you cannot see, in service of people — a donor, a patient, a student, a customer — who will never know or care which model drafted the words. Your part is the faithful part: choose well, verify honestly, keep the human where the judgment lives, and put your name on the result. The establishing — whether it lasts, whether it matters — was never finally yours to guarantee. So plan diligently and hold it with an open hand. That is what it means to build for a future you cannot see: not anxious, not arrogant, but faithful. Do your work as unto the Lord, and commit the rest to Him.

See you at the presentation — and then, go be the professional this whole course was for.