A Reader’s Companion

Depth at Work: Twelve Scenarios

The situations this week will actually throw at you — and the depth-preserving move in each.

The chapters argued the principles; the field guide gathered the protocols. This section is the layer between: twelve ordinary workplace moments where the depth deficit is quietly decided, each with the specific move that preserves the capability while keeping the efficiency. None takes more than minutes. All of them compound.

1. The meeting where AI takes the notes

The transcription is a gift — take it. The trap is what it licenses: attention half-spent, because ‘the summary will catch it.’ The summary catches the words; it misses the hesitation before the yes, the glance between two colleagues, the topic everyone steered around — the actual information in most meetings. The move: let the machine keep the record and give the room your full attention precisely because you no longer need to write. Afterward, before reading the summary, note the three things you observed that no transcript holds. That note is usually worth more than the minutes.

2. The performance review you’re tempted to have AI draft

A review is not a document; it is a judgment about a human being you are responsible for developing — and the drafting is where the judgment forms. Write the three sentences that matter yourself first: what this person does that they don’t know is valuable, where they are stuck, what would change their next year. Then let AI structure and soften around that spine. If you cannot write those three sentences, the problem is not drafting — you do not yet know your own report, and no model can know them for you.

3. The RFP due in forty-eight hours

Deadline pressure is where prompt-first feels mandatory and costs most, because RFPs are won in the two sections a machine cannot write: the reading of what the client is actually anxious about, and the claim only your firm can truthfully make. Thirty minutes on those two, by hand, before the tools inflate the rest. The boilerplate can be entirely delegated — it was always boilerplate. The two human sections are why you will or won’t be shortlisted.

4. Onboarding a junior on an AI-first team

The kindest thing you can do for a new professional is refuse to let their first year be prompt-supervised output. Give them the position brief habit on day one: their thinking first, the tool second, your review of both. It will make their first month slower and their fifth year possible — because the alternative, documented in Chapter 1, is a fourth-year professional with three years of outputs and no formed judgment, and by then you will both own the problem.

5. The incident postmortem

The AI can assemble the timeline in minutes — delegate that fully. The postmortem’s value is the twenty minutes after: the room asking, without the document open, ‘what did we each believe at the time that turned out wrong?’ Systems fail on timelines; teams fail on beliefs. The machine reconstructs the first. Only honest human conversation reaches the second, and it is the only part that prevents the sequel.

6. Preparing for the job interview

Use AI ruthlessly for the research layer — the company’s filings, the interviewer’s background, the likely questions. Then close the laptop for the only preparation that transfers into the room: saying your own answers out loud, badly, then better. An interview is a live-defense exam — Chapter 11’s format applied to your career — and reading beautiful prepared answers builds exactly none of the capability the room will test.

7. The salary negotiation

Have the AI build your market data and rehearse the counterarguments — it is excellent opposition research. But decide your actual number, and the reasons you believe it, before the tools speak, and write both down. Negotiation under pressure runs on conviction, and conviction cannot be borrowed; the moment the other side probes one layer beneath a script, only positions you genuinely hold will stand. Machine for the evidence. You for the resolve.

8. The conference talk

Slides, delegate. Structure, negotiate with the machine. But the talk itself — the twenty minutes of spoken conviction — must be built the way voice is always built: written in your own words, said aloud, cut, said again. An audience can feel generated language within ninety seconds; it is the byline test running live on two hundred faces. The speakers this book’s counter-movement produces are compelling precisely because what they say was earned before it was said.

9. The code review — or any review of machine-assisted work

The junior’s pull request was Copilot-assisted; so is everyone’s. The review question is no longer ‘did you write this?’ but ‘can you defend this?’ — walk me through why this approach, what breaks it, what you rejected. Review the understanding, not the authorship. It keeps the productivity and catches the one thing that matters: whether a human on the team can still reason about the system they are shipping.

10. The customer complaint that AI can answer instantly

The templated apology resolves the ticket and teaches nothing. Before sending it, one question held for sixty seconds: what is this complaint evidence of? The angriest customers are the cheapest consultants you will ever have, and the pattern across ten of them is product strategy — visible only to someone actually reading. Let the machine draft the response. Keep the reading, and the noticing, for yourself.

11. The language that isn’t your first

For professionals working in a second language, AI polish is a genuine equalizer — use it without apology. Preserve one thing: draft your meaning first, in whatever imperfect words, and let the tool refine expression rather than generate substance. Polished language carrying your thinking is assistance. Fluent language replacing it is the deficit wearing its most sympathetic disguise — and in the meeting, where no tool can whisper, only the first kind holds up.

12. The Friday afternoon with nothing urgent

The most consequential scenario is the emptiest one. The efficient professional now regularly hits pockets the tools have cleared — and the default fills them with more volume, more inbox, more feed. The move: one pre-decided depth investment that lives on a sticky note for exactly this moment — the hard problem you never reach, the paper in your field, the unassisted practice task, the call to your mentor. Empty hours are the dividend this entire book is about. The professionals who invest them differently are, a decade later, different people.

None of these moves costs more than minutes. All of them are decided in the moment nobody is watching — which is exactly where the two kinds of professionals diverge.