Project P6 — Inbox & Meeting Makeover

“Whatever you do, work heartily, as for the Lord and not for men,” — Colossians 3:23 (ESV)

Chapter: 6 — Everyday Productivity I: Words, Email, and Meetings Due: End of Week 6 Submit: A link to a shared folder (Google Drive / OneDrive / Dropbox) or a single combined PDF. Non-technical — no GitHub, no code. See Appendix A. Allowed tools: Any general assistant (ChatGPT, Claude, or Gemini) plus any meeting/notes tool you like. (Tool reference: Appendix B.) AI policy (Phase 1): AI drafts and reorganizes; you verify every commitment and every fact before it goes out. The provided materials are fictional and safe. Do not paste real client email, real names, or confidential material into a consumer tool — read Appendix C.

The Setup

Meet Jordan Ellis, office manager at Brightpath Learning Co., a small tutoring company. Jordan’s Monday is a swamp: seventeen unread messages by 8:14 AM — an upset parent demanding a call today, two overdue-invoice emails that don’t quite match, a catering quote with a deadline — plus the raw machine transcript of Thursday’s staff meeting, whose notes Jordan promised “by end of day” and still hasn’t sent. Priya’s first email of the morning is asking for them. By Tuesday, two things always slip — a commitment buried in an email thread, and an action item somebody agreed to in the meeting that nobody wrote down. Then it’s Jordan’s fault.

This week you become Jordan’s system. You’ll use AI to triage a messy inbox and turn a rambling meeting transcript into structured notes with owned action items — a name and a due date on every commitment. But here’s the discipline that separates a professional from a button-pusher: AI can propose an action item, but only a human can confirm that someone actually agreed to it. You’ll verify every one.

Setup (the starter)

This chapter ships three starter files in code/:

• code/meeting-transcript.txt — a realistic, messy staff-meeting transcript (interruptions, tangents, half-decisions). Your raw material.
• code/inbox.txt — the plain-text export of Jordan’s Monday-morning inbox: seventeen messages to triage. Some matter a lot, some are noise, and one or two are traps.
• code/action-items-template.csv — the columns your action-item table must fill: Owner, Action, Due Date, Source Quote (from transcript), Depends On / Condition, Verified By Hand (Y/N), and Notes. It includes one worked example row to show the format.

Copy all three into your project folder.

Learning Targets

You will demonstrate that you can:

• Triage with AI — turn a chaotic inbox into a prioritized, categorized action list without losing anything that mattered.
• Structure a meeting — convert a messy transcript into clean notes plus action items with an owner and a due date each.
• Verify every commitment — check each AI-extracted action item back against the transcript, catching the ones the AI invented or misattributed.
• Keep your voice — draft replies with AI that still sound like a real person at Brightpath, not like a chatbot.

Normal Tier

Goal: Triage the inbox, structure the meeting, and produce a verified action-item list — with the human in the loop on every commitment.

Required features

Produce a folder named p6-makeover containing:

1. inbox-triage.docx (or a table). Feed code/inbox.txt to an assistant and produce a triage: each message categorized (reply now / delegate / schedule / archive), with a one-line reason. Then verify — flag at least one item the AI mis-prioritized or missed.
2. meeting-notes.docx. From code/meeting-transcript.txt, produce clean structured notes: decisions made, topics discussed, and open questions. It should be readable by someone who wasn’t there.
3. action-items.csv — filled from the template. Every action item from the meeting with an owner, an action, and a due date, plus the exact source quote from the transcript that supports it, any condition it depends on, and a Verified By Hand mark. The source-quote column is the verification: if you can’t point to the line where someone agreed, the item doesn’t belong.
4. The verification pass. AI extraction of action items is where hallucination bites — models invent owners and deadlines that were never said. Find and remove at least one fabricated or misattributed action item, and note what tipped you off.
5. Two AI-assisted replies — in your voice. Draft replies to two inbox messages with AI, then edit them so they sound like a real Brightpath employee. Keep both the raw AI draft and your edited version, and note what you changed and why.
6. The AI usage note. One honest line: what you asked, what the AI got wrong (especially in the action items), what you verified.

Normal-tier rubric (out of 100)

Medium Tier (+up to 25% extra credit)

Pick one or more:

1. Prompt for the source line. Rewrite your extraction prompt so the AI is required to quote the transcript line that justifies each action item. Compare hallucination rates before and after. (You’re building the Week-12 skill early.)
2. Two tools, one meeting. Run the transcript through a general assistant and a dedicated meeting-notes tool. Which produced better action items? Note the difference.
3. The follow-up email. Have AI draft the post-meeting “here’s what we agreed” email from your verified notes, then edit it. Include what you changed.

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

This is the judgment the machine cannot make for you.

Write process-memo.docx — a one-page memo to Jordan’s boss proposing a standing weekly process for AI-assisted inbox and meeting handling at Brightpath. You must:

• Draw the line. Name exactly which steps AI does (draft, categorize, extract) and which a human must always do (confirm commitments, send anything to a parent, decide priorities). Tie it to the spine rule.
• Name the failure you’re preventing — the invented action item, the reply sent without reading, the parent message a bot got wrong — and the control that stops it.
• Make it real. It should be something a busy tutoring company could actually adopt Monday.

A memo that says “let AI handle the inbox” fails — you just watched AI invent a commitment. The grade is in showing exactly where the human stays in the loop, and why.

Submission

One shared folder or a combined PDF; submit the link per Appendix A:

p6-makeover/
  inbox-triage.docx    (or a table)
  meeting-notes.docx
  action-items.csv
  replies/             — the two before/after AI-assisted replies
  process-memo.docx    (Hard tier)

Do not use real student or parent information, real staff names, or any confidential material — the fictional Brightpath materials exist so you never have to.

Hints (Read Before You Begin)

• Make “cite the source line” a rule, not a wish. The single best defense against invented action items is forcing the AI to quote where the commitment was made — then checking it yourself.
• A owner without a name is not an owner. “Someone will handle it” is how things fall through. Every item gets a person.
• Read before you send. The whole discipline of the week is that a fast draft is not a sent message until a human has read it.
• Keep your voice. The goal isn’t AI-perfect prose; it’s a real reply that sounds like Brightpath.

What Mastery Looks Like (Beyond the Rubric)

A mastered submission is one where I trust your action-item list — because every row points to a line someone actually said, and you caught the one the AI made up. Your notes are readable by someone who missed the meeting. Your replies sound human. And your Hard-tier memo draws a clean, defensible line between what the tool does and what Jordan must never delegate. The speed is nice. The verified speed is the job.

Coach’s Note — AI will happily hand you a beautiful set of action items, two of which nobody agreed to. If you forward that list, you didn’t save time — you manufactured a problem with a confident face on it. The work heartily part isn’t slower; it’s just honest. Draft fast, verify every commitment, sign it yourself.

When You’re Done (a short checklist)

• Inbox triaged; at least one AI mistake flagged.
• Meeting notes clean and readable by an absentee.
• Action items complete with owner, due date, and a real source line each.
• A fabricated/misattributed action item found and removed.
• Two AI replies edited into your voice; before/after kept.
• Honest AI usage note; no real/confidential data.
• (Hard) process-memo.docx draws the human-in-the-loop line.

A theological footnote. “Whatever you do, work heartily, as for the Lord and not for men.” The temptation of a productivity tool is exactly the opposite spirit — to do the least, to let the machine’s confident output stand in for care, to forward a list you never read because it looked finished. Paul’s word cuts against that. Working heartily doesn’t mean working slowly or refusing good tools; it means the quality of your attention doesn’t drop just because a draft appeared for free. Jordan’s colleagues, the parents waiting on a call, the students whose workbooks get reordered or don’t, are real people served by ordinary, faithful attention to small things. The AI can carry the typing. It cannot carry the care. That part is yours, and it is worth doing as unto the Lord.

See you next week.