Chapter 6 — Everyday Productivity I: Words, Email, and Meetings

“The palest ink is better than the best memory.” — Chinese proverb

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

Why This Matters

For five weeks we studied the machine. You learned what a model is, how the labs ladder their families, how to choose a tier, how to talk to one, and how to reach for the right tool instead of the only one you knew. That was the classroom. This week the whistle blows and we walk onto the field, because starting now the course is about your actual Tuesday — the inbox that is already at forty unread when you sit down, the meeting you half-remember, the report due Friday that you have not started.

Here is the plain truth about knowledge work in 2026: most of it is words. You read words, you weigh them, you write more words back. Email, memos, notes, replies, summaries, the recap nobody wrote so nobody remembers what was decided. That is the raw material of the professional day, and it is exactly the material a large language model is built to handle. Which means AI is about to change your Tuesday more than it changes almost anything else — and it will change it whether or not you are ready.

So let us be ready. This chapter is the first of two on everyday productivity. This week: the daily work of words — drafting and improving writing, triaging a full inbox, replying without dread, brainstorming with a partner who never gets tired, and turning a messy meeting transcript into structured notes with owned action items. Next week (Chapter 7) we take on research, long documents, and data. The two chapters are a pair, and the pair is the most immediately useful thing in this book. You will use it the day after you read it.

But — and you knew a but was coming — the same speed that saves your Tuesday can quietly wreck it. A tool that drafts a reply in four seconds also lets you send a reply in four seconds that you never actually read. A tool that summarizes a meeting will confidently list an action item that nobody agreed to, assign it to the wrong person, and invent a deadline that was never spoken. The danger this week is not that the AI is too weak to help. It is that it is just good enough to make you sloppy.

That is why the spine of this whole book matters more here than anywhere:

You choose the tool. You own the verdict. AI drafts, generates, and accelerates; the professional decides, verifies, and is accountable. AI is an assistant, not an authority.

Restate it in this week’s terms and it becomes a work ethic: the AI writes the draft, but the words go out under your name. When the client reads that email, they are not reading “the AI’s email.” They are reading yours. When your boss acts on those meeting notes, they are trusting your notes. The tool did the typing. You did the sending — and the sending is the part that carries your name, your judgment, and your reputation.

That brings us to this week’s question, and I want you to sit with it, not skim it: what does it mean to work heartily, as for the Lord, when the tool makes it so easy to be sloppy? The proverb at the top and the verse beneath it are asking the same thing from two directions. One says write it down — memory fails. The other says do the work with your whole heart, as unto God. Both are about caring enough to do the thing well when it would be easier to do it fast and fake. Hold that. We will earn it in 6.10.

Coach’s Note — If you are new to the Litman library, here is the house rule. Learning is a sport. The keyboard is the gym. You do not get better at writing with AI by reading about it — you get better by drafting, catching a hallucinated deadline, fixing it, and feeling the difference. Every chapter ends with Reps and a Project. This one especially. Show up.

6.1 — The Day Is Made of Words

Sit at almost any professional’s desk and watch what actually happens. A nurse charts a shift and writes a handoff note. A teacher answers parent emails, drafts a newsletter, and writes report-card comments. A small-business owner replies to a supplier, posts to social, and writes up a quote. A marketer drafts copy, an office manager writes the meeting recap, an account rep answers the same five questions in five slightly different emails. Different jobs, same substrate: read, judge, write.

For all of history that substrate had a fixed cost. Writing a good paragraph took the time it took. The blank page was a tax everyone paid. What changed in 2026 is that a machine can now pay the first installment of that tax for you — it can produce a plausible first draft of nearly any workaday document in seconds. Not a finished document. A first draft. The distinction is the entire chapter.

Understand what that shifts. It does not shift the judgment — whether the message is true, kind, appropriate, and actually says what you mean. It shifts the starting line. You used to start at a blank page; now you start at a rough draft. And starting at a rough draft is a genuinely different, faster, less-dreadful way to work — if you remember that a draft is a starting line and not a finish line. The professional who thrives this year is not the one who generates the most words. It is the one who still reads every word before it leaves their hands.

Coach’s Note — Notice the two tasks hiding in “using AI to write”: correcting a draft and trusting a draft. They feel identical — you’re looking at the same screen of text — and they are opposite. One ends with you owning the words. The other ends with the machine owning you. This whole chapter is about staying on the correcting side of that line.

6.2 — The Draft-Then-Verify Habit

If you take one habit out of this chapter, take this one. Everything else is application.

Draft, then verify. Let the AI produce the first version. Then you — a human with the context, the stakes, and the accountability — go through it and check the things a machine cannot be trusted to get right: the facts, the names, the numbers, the dates, the tone, and whether it says what you actually mean. Only then does it go out.

Say it as a loop you run every time:

1. Frame the task for the tool — audience, goal, tone, what it needs to know (this is the prompting you learned in Chapter 4).
2. Draft — let it generate.
3. Read every word. Not skim. Read.
4. Verify the load-bearing parts — every claim, name, figure, and commitment against something you actually know or can check.
5. Make it yours — fix the voice, cut what’s wrong, add what it missed.
6. Send — and own it.

Steps 3 and 4 are the ones people skip, and they are the whole job. The AI is fluent, and fluency is disarming — a wrong answer arrives in the same confident, well-formatted prose as a right one. You cannot tell truth from fiction by tone, because the tool has exactly one tone: certain. (We will hit this head-on in Chapter 8 when we study hallucination directly.) So you do not verify by feel. You verify by checking.

Coach’s Note — A hallucination is a confident, fluent, plausible statement that is simply false — not a glitch you can patch out, but a property of how these tools generate text. For the daily work of words, that means: assume the shape is right and the specifics need checking. The AI is usually right about how a polite decline reads and often wrong about the client’s name, the meeting date, or the figure you quoted.

Here is the sharpest version of the rule, and I want it on a sticky note above your monitor: never send anything you did not read. Not the email the AI drafted. Not the summary it produced. Not the “quick reply” it suggested. If your eyes did not pass over every word with your judgment engaged, you did not write it and you are not ready to sign it. Autofill is not authorship. A signature is a claim: I read this, I mean it, hold me to it.

6.3 — Keeping Your Own Voice

There is a cost to letting a machine draft your words, and it is not the one people worry about. The worry is plagiarism or cheating. The real cost is blander: everything starts to sound the same. Default AI prose has a texture — smooth, agreeable, slightly inflated, fond of “delve,” “leverage,” “in today’s fast-paced world,” and tidy rules of three. It is competent and it is nobody’s. Send enough of it and your colleagues stop hearing you.

Your voice is an asset. It is how people know a message is really from you, how trust accrues, how a “thanks, this made my day” reads different from a form letter. Do not trade it away for speed. Here is how you keep it while still using the tool:

• Give it a sample of your writing. Paste two or three of your own real emails and say “match this voice — direct, warm, no corporate filler.” The tool is a chameleon; show it the color.
• Draft in the tool, finish in your own words. Use the AI’s draft as scaffolding, then rewrite the opening and closing lines yourself — those are where voice lives most.
• Cut the tells. Delete the throat-clearing intro and the inflated adjectives. If a sentence sounds like a press release, it is not yours.
• Keep your quirks. If you always open with the person’s first name and no “Dear,” keep doing that. Consistency is voice.
• Read it aloud. If you would never say it out loud to that person, do not send it in writing. Your ear is a better editor than the tool.

The goal is not to hide that you used AI. (Disclosure is its own question — see Appendix C.) The goal is that the finished words are true to you, because they are going out under your name and they should sound like the person whose name that is.

6.4 — Email Triage: Reading the Whole Inbox Fast

Now the inbox, which for most professionals is where the day goes to die. We ship a realistic one to practice on: code/inbox.txt — seventeen unread messages on a Monday morning, the way they actually arrive. Open it. It is Jordan’s inbox; Jordan runs the office at a small tutoring company, and Jordan is you this week.

There are two distinct jobs an AI can help with, and you should keep them separate in your head:

Job one: triage. “Here are seventeen email subject lines and one-line previews. Sort them into: needs a reply today, needs a reply this week, FYI only, and ignore/delete. Flag anything that looks urgent or risky.” This is a classification task — cheap, fast, well within a small model’s reach (Haiku-tier work, in the language of Chapter 3). It turns forty unread into a ranked list in seconds, and that alone can hand you your morning back.

But — and here is the judgment the machine does not have — the AI does not know your world. It does not know that “Re: Eli’s tutoring — still waiting” is the message that will become a crisis if it sits. It does not know which parent is a flight risk and which sender is a vendor you can safely ignore for a week. It sorts by the surface of the text; you sort by the stakes, and the stakes live in your head, not in the subject line. So AI triage is a first pass, not a verdict. You skim its ranking and override it with what you know.

Job two: spotting the traps. Look carefully at code/inbox.txt and you will find two messages about the same overdue invoice — one from ar@copyworld-print.example and one from a different sender, billing@copyworld-print-invoices.example, demanding you wire money to new bank details within 24 hours. That second one is how invoice fraud works. An AI triage tool might well rank the urgent-sounding one higher, because it sounds urgent — which is exactly backwards. This is why triage output is a suggestion you inspect, never a queue you act on blindly. Money movement and anything that smells off get verified out of band — you call the vendor at a number you already have, not the number in the email. (We will meet the deepfake version of this problem later; the defense is the same: a known channel, not the one the message hands you.)

Coach’s Note — The dangerous email is rarely the one that looks dangerous. It is the ordinary-looking one you skimmed at 8:14 AM and half-answered from your phone. AI triage is genuinely useful for beating the volume — but the moment it touches money, personal data, or a promise, the human slows down and verifies. Speed on the noise; care on the signal.

One more thing you must not do: do not paste confidential or personal information into a random consumer chatbot to triage it. Client names, patient details, student records, anything private — that is a hard line, and it is why Job one above works on subject lines and previews, not on the private contents of the messages. Concordia licenses BoodleBox partly to change this calculus: because it is FERPA-compliant and does not train on your data (as of 2026), it is the appropriate home for real Concordia work content when a public tool never would be — the full argument waits in Week 14. Even so, the reflex here stays the same: when in doubt, keep it out. Appendix C is the full rulebook; internalize the one-line version now.

6.5 — Drafting Replies Worth Sending

Triage tells you what to answer. Now you have to answer it. This is where AI earns its keep on an ordinary day — and where “never send what you didn’t read” gets tested hardest, because a good draft is so tempting to just fire off.

The move is simple. Take a message that needs a reply, give the tool the context and the outcome you want, and let it draft. For the flagged Henderson message in the inbox, you might frame it: “A parent is upset that her son was placed with a tutor who isn’t certified in our reading program. She wants a phone call today. Draft a short, warm, non-defensive email acknowledging her concern, confirming I’ll call this afternoon, and not admitting fault or making promises I can’t keep. My voice: direct, human, no corporate filler.” Notice everything you supplied — the situation, the tone, the boundaries (“no promises I can’t keep”). That framing is the Chapter 4 skill, and it is what separates a usable draft from generic mush.

Then you verify, and for a reply the checklist is specific:

• Is every fact true? Did she actually ask for a call? Is the tutor really uncertified, or is that her claim? The draft will state things as fact that are only alleged — fix them.
• Is the name right? (More on this in a second. Names are where these tools quietly fail.)
• Does it promise anything I can’t keep? AI drafts love to be reassuring. “We’ll fix this immediately” is a commitment you may not be able to make.
• Is the tone right for this person? Warm, but not groveling. Professional, but not cold.
• Would I be comfortable if this were forwarded to my boss? Because it might be.

A special word on the decline and the hard message — the “no,” the bad-news email, the pushback. These are the ones people most want to hand to the machine because they are uncomfortable to write. AI is genuinely good at giving you a calm, professional scaffold for a message you are too annoyed to write cleanly. Use it for exactly that. But the judgment about how firm to be, what to concede, what to hold — that is yours, because you are the one who has to live with the relationship afterward. The tool can find you the polite words. It cannot decide how much to give away.

6.6 — Reports, Memos, and the Blank Page

The blank page is a tax, and the report is where you pay the most of it. Here the AI’s honest value is at its clearest: it demolishes the blank page. You are no longer staring at nothing; you are editing a rough draft, which is a far easier cognitive task.

A reliable workflow for any longer document:

1. Dump your raw material in. Bullet points, half-sentences, the numbers, the three things that happened this quarter. Messy is fine — the messier the input, the more the tool is actually helping.
2. Ask for structure first, prose second. “Turn these notes into an outline for a one-page status report for my manager.” Approve the skeleton before you let it write paragraphs. Fixing an outline is cheap; rewriting three finished paragraphs is not.
3. Generate section by section. Long single-shot documents drift and pad. Shorter asks stay tighter and are easier to verify.
4. Verify every number and claim. This is a report — someone will make a decision from it. Every figure traces back to a real source you checked. The AI did not look up your Q2 revenue; do not let it invent one.
5. Cut it by a third. AI prose is padded by default. The best edit you will make is deletion. (There is an old line — “I would have written a shorter letter, but I did not have the time” — that is usually attributed to Pascal, though the attribution is disputed. True either way: short is the harder, better work, and now you have the time.)

Coach’s Note — The single most valuable thing the tool does for long documents is not the writing — it is the structuring. Handing a machine a pile of messy notes and getting back a sensible outline is worth more than any polished paragraph, because the outline is where your thinking gets organized. Let it structure. Then you write the parts that carry weight.

The trap in reports is the fabricated specific. The document reads beautifully and contains a statistic, a citation, or a dollar figure that the model invented to make the paragraph flow. It is not lying on purpose; it is completing a pattern. Fluent prose wants a number in that sentence, so it supplies one. Your job is to treat every specific as unverified until you have checked it. We drill exactly this — grounding claims in a real source — next week in Chapter 7.

6.7 — Brainstorming With a Thinking Partner

Not every use of AI ends in a document. One of the best uses ends in your own clearer thinking. The tool is a tireless, judgment-free, always-available brainstorming partner, and for a lot of professionals that is where it quietly changes the most.

Where it shines:

• Quantity on demand. “Give me twenty subject lines for the fall enrollment email.” You will keep two, but you would never have generated twenty alone, and one of the eighteen you reject will spark the one you write yourself.
• The devil’s advocate you can summon. “I’m about to tell a parent X. Argue the other side — what am I missing? How could this land badly?” An AI will poke holes in your plan without ego and without gossiping about you afterward. That is a rare and useful thing.
• The rubber duck. Explaining your half-formed idea to the tool, in writing, forces you to make it make sense — and half the time you solve it yourself mid-sentence.
• Reframing. “Say this three ways: for a nervous parent, for my boss, for a colleague.” Seeing the same idea in three registers sharpens which one you actually mean.

The discipline here is subtle but real: brainstorm with it, don’t outsource the thinking to it. The ideas that matter are still yours to choose, judge, and own. The tool multiplies your options; it does not get to pick. A brainstorm where you accept the AI’s first list wholesale is not a brainstorm — it is a surrender with extra steps. Use it to think more, not to think less.

6.8 — From a Messy Transcript to Structured Notes

Now the marquee skill of the week, and the one this chapter’s lab and project are built around: turning a meeting into something useful.

You know the failure mode. A thirty-minute meeting happens, decisions get made, six people leave with six different memories of what was decided, and a week later nobody can say who owns what. The proverb at the top of this chapter is the whole diagnosis: the palest ink is better than the best memory. Meetings evaporate. Notes endure. And writing good notes during a meeting is nearly impossible, because you cannot both participate and transcribe.

This is a task AI is genuinely, powerfully good at — with a large, sharp caveat. Paste a raw transcript into BoodleBox — the platform Concordia licenses (sign in with your Concordia account at box.boodle.ai) — or into a dedicated meeting tool (Otter, Fireflies, Fellow, and others in Appendix B), and it will produce a clean summary, a decision list, and a set of action items in seconds. What used to take you twenty minutes after the meeting takes ten. (Off-campus or without a license, any general assistant runs the same play.)

Open the raw transcript we ship: code/meeting-transcript.txt. It is deliberately, realistically messy — auto-transcribed, full of “um,” speaker labels that are just “Speaker 1,” crosstalk, half-finished sentences, and decisions buried in the noise. This is what a real machine transcript looks like. Feed it to an AI and ask for a structured summary and you will get something impressively clean back.

And then you check it, because clean is not the same as correct. Read the transcript yourself and you will find the traps that a summarizer routinely gets wrong:

• The transcript mis-hears a name — “Mrs. Anderson — sorry, Henderson” — and a careless summary can carry the wrong name forward.
• Two people disagree on the open-house date (the 24th vs the 25th) before it’s resolved to the 24th. A summary might grab the wrong one.
• One “action item” — sending the flyer to the printer — is explicitly put on hold thirty seconds after it is proposed. A naive extraction lists it as a live task. It is not.

Structure the summary into three buckets and it becomes usable: Decisions (what was settled), Discussion (context worth keeping), and Action Items (who does what by when — the next section). The AI produces the buckets. You verify the contents. That division of labor is the whole skill.

In BoodleBox — Do this week’s drafting in the tool Concordia already pays for: start a chat, paste the transcript, and ask BoodleBot for the three buckets. Because BoodleBox is FERPA-compliant and does not train on your data (as of 2026), it is the right place for real meeting content — not a random chatbot. And when the notes are ready to hand off, don’t just email a static copy. Open a Box — a shared chat that holds both people (your teammates) and bots — so the team can read the notes, ask the bot a follow-up, and see the source transcript in one place; then group related Boxes into a Folder so there’s one home for the team’s meeting notes. Off-campus, the same steps work in any assistant plus a shared drive.

6.9 — Owned Action Items: Who, What, By When

An action item without an owner is a wish. This is the line where meeting notes become management, and it is the exact spot where AI extraction most needs a human.

A real action item has three parts, always: an owner (a named person, not “the team”), a task (specific enough to be done), and a due date (a real date, not “soon”). Miss any one and the item quietly dies. AI is good at drafting these from a transcript — it is genuinely handy at spotting “I can have that done by Tuesday” and turning it into a row. But it makes three characteristic errors, and you must catch all three by hand:

1. Wrong or missing owner. The transcript says “someone should probably follow up” and the AI assigns it to whoever spoke last, or to nobody. Ownership is a decision, and if the meeting didn’t make it, the notes can’t invent it — they flag it: owner TBD, confirm.
2. Invented or vague dates. “End of week,” “before the open house,” “today or tomorrow” — the AI will often either drop these or harden them into a specific date that was never said. You pin the real date, and where the meeting was vague, you write down that it was vague.
3. Phantom items — commitments that were revoked, conditional, or never agreed. The flyer-to-the-printer task in our transcript is the canonical example: proposed, then killed. A commitment that depends on an approval (“send the email pending Priya’s review”) is not the same as a green light, and a good notes-taker records the condition, not just the task.

We give you a scaffold for this exact discipline: code/action-items-template.csv. Its columns are the checklist — Owner, Action, Due Date, the source quote from the transcript, the condition it depends on, a Verified By Hand (Y/N) column, and Notes. That “source quote” column is the heart of it: every action item points back to the words that created it. If you cannot quote the line where a commitment was actually made, it is not an action item — it is something the AI inferred, and inference is where the errors live.

Coach’s Note — Here is the rule that makes you the person people trust with the notes: no action item ships until you can point to the sentence that created it and name the human who owns it. The AI drafts the rows in seconds. You spend two minutes tracing each one back to a real quote and a real owner. Those two minutes are the entire value of the notes. Skip them and you have produced a confident, tidy document full of commitments nobody made — which is worse than no notes at all.

6.x — Interactive Lab: Transcript → Action Items

Below this chapter on the website you will find an interactive panel called Transcript → Action Items. Go use it now — it is not decoration, it is the rep that puts this week’s lesson in your hands.

The panel gives you a messy meeting transcript — the same kind of raw, auto-transcribed text as code/meeting-transcript.txt, full of filler, crosstalk, and buried decisions. You watch it get pulled apart into structured action items: owner, task, due date. Your job is not to admire the extraction. Your job is to audit it. None of the rows it produces survive unexamined — every one needs at least a fix, and your job is to catch each one’s flaw. Some have the wrong owner. Some have a due date that was never said, or a date the meeting corrected later. At least one is a “commitment” that was put on hold and should not be on the list at all.

For each extracted item you decide: keep it, fix it, or kill it — and the lab shows you the transcript line the item came from, so you can check the AI’s work against the actual words. That “trace it back to the source line” motion is precisely the habit from 6.9, and the lab is where it goes from a paragraph you read to a reflex you own.

Run it twice. The first time, trust the extraction and see how many bad items you would have sent. The second time, audit every row against its source line before you accept it. The gap between those two runs — the phantom tasks, the wrong owners, the invented dates you caught on the second pass — is the skill this whole chapter is teaching. That gap is the difference between notes people trust and notes that quietly cause a mess.

In BoodleBox — The panel is the warm-up; run the real rep in BoodleBox. Start a chat, paste code/meeting-transcript.txt, and ask for action items with an owner, task, and due date — then audit each row against the transcript exactly as the panel taught you. If an extraction looks shaky, get a second opinion without leaving the chat: type @ to open the bot picker and add a second model, then compare the two side by side (as of 2026 the picker’s labels shift, so go by the action — bring another model into the same conversation). No BoodleBox on hand? Any general assistant runs the same rep.

6.10 — Work Heartily, as for the Lord

Now the week’s question, given its full weight. What does it mean to work heartily, as for the Lord, when the tool makes it so easy to be sloppy?

Start with the verse. “Whatever you do, work heartily, as for the Lord and not for men” (Colossians 3:23, ESV). Paul is writing to ordinary people about ordinary work — in the original context, to servants about the daily labor of a household. He is not addressing monks or apostles. He is addressing the person doing the unglamorous, repetitive, watched-by-no-one work that fills a normal day: the email, the notes, the report. And he says: do that — the small, the tedious, the invisible — heartily, with your whole self, as for the Lord. The very next verse names the reason: you are, finally, “serving the Lord Christ” (Colossians 3:24, ESV). The audience for your work is not only your boss. It is God.

Now feel how sharply that cuts against what the tool tempts you toward. The temptation of AI in daily work is not laziness in the obvious sense — you are still producing, still shipping, the inbox is still getting cleared. The temptation is subtler and more corrosive: to produce without caring. To let the machine draft it, glance at it, and send it — technically done, actually abandoned. To hand off not just the typing but the attention. And attention is exactly what “heartily” is. To work heartily is to bring your whole care to the thing, even when — especially when — no one would catch you if you didn’t.

This is where the Christian sees something a productivity book cannot say. The world’s reason to do good work is that someone is watching — the boss, the client, the review. But the tool makes it trivial to look like good work without doing it, and much of the time no human will ever know the difference between an email you read and one you didn’t. The world’s motive fails at exactly the moment the tool makes sloppiness invisible. Paul’s motive does not fail there, because the audience he names cannot be fooled by a fluent draft. You work heartily “as for the Lord” precisely because the Lord sees the email you sent without reading, the action item you assigned to the wrong person because you couldn’t be bothered to check, the number you let the machine invent. Coram Deo — before the face of God — there is no such thing as unwatched work.

And this is not grim. It is dignifying. It means the tedious Tuesday email is not beneath your full care — it is a place to serve your neighbor and honor God, done well. LCMS theology calls this vocation: God works through your ordinary calling to care for the people your work touches. The parent waiting on the reply, the colleague trusting your notes, the client reading your report — Christ is present in that neighbor, and your diligent, honest, read-before-you-send work is how you love them. The AI is a fine servant in that calling. It drafts; you care. It is fast; you are faithful. To use it well is not to work less. It is to be freed from the drudgery of the blank page so you can spend your care where care belongs: on the truth of the words, the good of the reader, and the honesty of the thing that goes out under your name.

So: work heartily. Let the tool draft. Then read every word, verify every fact, keep your own voice, and send it as unto the Lord — not because a human is watching, but because the One who matters most always is.

6.11 — Common Pitfalls

Pitfall: Sending AI output you never actually read. Example: You ask for a reply to an upset parent, skim the first two lines, and hit send — and paragraph three promised a full refund you never intended to offer. Fix: Never send anything you did not read. Every word, every time. Autofill is not authorship; a signature is a claim that you read it and mean it.

Pitfall: Trusting extracted action items without tracing them to the source. Example: The AI lists “send the flyer to the printer by Thursday” as a task — but in the meeting that was explicitly put on hold pending budget approval, and now the flyer gets printed against instructions. Fix: No action item ships until you can quote the transcript line that created it and name a real owner. Use the source quote column in code/action-items-template.csv.

Pitfall: Letting the machine flatten your voice into corporate mush. Example: Every email you send now opens with “In today’s fast-paced environment” and your colleagues stop being able to tell your messages from a template. Fix: Give the tool a sample of your real writing, finish the opening and closing in your own words, and cut the tells. Read it aloud — if you wouldn’t say it, don’t send it.

Pitfall: Acting on AI triage of an inbox without checking the risky items yourself. Example: The triage ranks an “URGENT: wire payment to new bank details” email near the top, you treat “top” as “legitimate,” and you nearly wire money to a fraudster. Fix: Triage is a first pass, not a verdict. Anything touching money, personal data, or a promise gets verified out of band — call a number you already have, never the one in the message.

Pitfall: Pasting confidential or personal information into a consumer chatbot. Example: You paste a full email thread containing a client’s private details into a public AI tool to “summarize it,” and that data is now outside your control. Fix: Triage and summarize on subject lines and non-sensitive text. Client, patient, and student data stay out of consumer tools — when in doubt, keep it out. See Appendix C.

Pitfall: Trusting a fabricated specific because the paragraph reads well. Example: Your AI-drafted report cites “a 34% increase in retention” — a number the model invented to complete the sentence — and your manager repeats it to their boss. Fix: Treat every number, name, date, and citation as unverified until you personally check it against a real source. Fluent is not the same as true.

6.12 — Reps

The work is in the exercises. The keyboard is the gym; this is where Week 6 gets into your hands. A preview of what is waiting:

• Rewrite a real email three ways and feel where the AI’s default voice is not yours — then bring it back to yours.
• Triage the shipped inbox code/inbox.txt by hand, then with AI, and find the two messages the AI would rank exactly backwards.
• Convert the messy transcript code/meeting-transcript.txt into structured notes, then hunt for the wrong name, the corrected date, and the phantom action item.
• Fill the action-items template code/action-items-template.csv and prove every row traces to a real quote and a real owner.
• Draft a hard “no” with the tool, then decide for yourself how firm to be — the judgment the machine cannot make.

This week’s AI policy for reps: work in BoodleBox first (sign in with your Concordia account at box.boodle.ai), or any public AI assistant off-campus — but every rep that uses it ends with an honest one-line AI usage note — what you asked, what it got wrong, and what you verified by hand. The human owns the verdict.

A short Check Your Reps quiz is embedded on this page, right under the chapter. Take it before you move on — five questions, straight from what you just read.

6.13 — This Week’s Project

Your project is P6 — “Inbox & Meeting Makeover,” specified in Project 6. You step into Jordan’s Monday: a week of email to triage and draft replies for, and a messy meeting transcript to turn into structured notes with owned, verified action items. You will use AI to accelerate all of it — and then verify every commitment by hand, because the notes go out under your name.

At a high level: Normal tier triages the inbox, drafts the replies, and produces verified meeting notes plus a clean action-item table. Medium tier adds your pick of a source-line-required notes prompt, a two-model comparison of the same meeting, or the AI-drafted follow-up email. Hard tier is a one-page memo — a judgment call an AI cannot make for you — recommending what your organization should and should not let AI touch in the daily work of words, and defending the line. That memo is where this week’s thesis gets graded.

6.14 — Coach’s Final Word

Here is what I want you to carry out of Week 6. The tool just took the most dreaded part of your day — the blank page, the full inbox, the meeting you have to reconstruct — and cut its cost to near zero. That is real, and you should take the gift. Your Tuesdays are about to get lighter.

But lighter is not the same as better, and the difference is entirely you. The AI made it effortless to produce words. It did nothing to make those words true, kind, right for the reader, or actually said in the meeting. All of that — the part that was always the real work — is still yours, and now it is more yours, because the tool has stripped away the busywork that used to hide behind. There is nowhere left to hide. When the words go out under your name, the machine’s speed is not a defense. Your reading, your verifying, your care — those are the job now.

So do the unglamorous, exact thing. Read every word before you send it. Trace every action item to the sentence that made it. Keep the number, the name, the date honest. Keep your own voice. Do it heartily — not because a boss is watching, and not because the tool would catch you, but because the work itself deserves it and the neighbor on the other end is real. The palest ink beats the best memory; the read email beats the sent-and-forgotten one; and work done as unto the Lord beats work done to merely look done, every single time.

Now go do the reps. The lab is waiting right below this page, the inbox and the transcript are in code/, and Project 6 is where it all comes together.

See you on Monday.

Up next: Read the exercises and do all of Week 6’s reps, then build Project 6 — Project P6: Inbox & Meeting Makeover. Set up your toolkit from Appendix A, keep the tool directory Appendix B open for meeting-notes tools, live by the rules in Appendix C, and check any unfamiliar term in Appendix D. Then Chapter 7 — Everyday Productivity II: Research, Documents, and Data.