Cognitive Transition · Book One

Chapter 6 · Thinking Better in an Automated Environment

“This isn’t a read-through. It’s an interrogation.”

Summary

A progress meeting, a report on the table, one question: so who decided this? Silence. The chapter starts there because that scene opens the practical part of the book, the part that stops looking at the tool and redesigns the human who governs it. It proposes separating four layers that work without AI kept fused in a single person, and placing the machine in the one position where it performs without risk, that of generator. It explains why a measure of deliberate friction protects more than it costs, and why, when producing is nearly free, judgment becomes the advantage people pay for. All of it takes shape in a five-phase protocol that puts human thinking before and after automated generation, and in a six-field hybrid decision template for keeping your own record.

Extended content

Liquid part: this text keeps growing after the printed edition.

Three lines before the first click

Tomorrow, before you open any tool, write three lines with your hypothesis about the first important task of the day. Note the time. When the model's answer arrives, set it beside yours and mark where they disagree and what changed your mind, if anything did. This is phase one with a stopwatch, and it measures one thing: how much of your final answer was yours before the other one arrived.

Irene and the workshop

Irene runs a small training consultancy and has to propose the program for a two-day workshop for a client's sales team. She writes down her hypothesis: that team doesn't need more sales technique, it needs to learn how to lose a negotiation well. She asks the model for eight program variants and gets eight tidy programs.

Then she makes the slip the chapter describes and asks which one is best. The model answers with confidence, and Irene reads the answer the way you read a colleague who missed the earlier meeting. The recommended variant includes three role-plays. She knows the sales director hates them after a previous experience, and that two of the attendees are coming in wanting to argue about something else. None of that was in the prompt, and none of it could have been.

She discards five variants, rescues from the third an exercise she wouldn't have thought of, writes in her own words why the program ends up as it does, and signs it. In the template she records the reason for the most important discard. Three months later, that line saves her an argument with another client with a similar profile.

Sorting your week and installing a gate

List the ten tasks where you used AI last week. Give each one a tier: trivial, significant, or critical. Look first at the ones you sorted without thinking. Then find the one you treat as trivial and that, if it goes wrong, someone other than you pays for. That is the mistake in the comfortable direction. Move it up a tier for a month and watch what changes in how you do it.

Find a point where an output leaves your hands without anyone else looking at it: the client email, the code going to the main branch, the quote. Install a single gate, the cheapest one you can think of, whether that's a one-hour wait, a second read by someone who didn't ask for the text, or three written questions.

Also find a control that already exists and that everyone skips. If they skip it every day, it's in the wrong place. Remove it or move it to where an error is expensive, because friction that gets ignored is worse than friction that's missing.

Five minutes of phase five

The phase almost everyone skips is the one that pays off most. After each significant decision, write four lines: what you asked for, what you discarded, which of the model's errors you spotted, and what you'd do differently next time. Five minutes, not one more. After a month you'll have a pile of records made by your own hand; read them in one sitting and underline whatever repeats. That underlining is closer to a manual for your trade with AI than any course you could buy, because it was written by someone who knows your mistakes.

Questions to work with the protocol

Which part of my last important decision could I explain without looking at the output?
How many times this week did I ask "which one is best," and how many times "give me ten"?
Who signs what leaves my desk, and do they know it?
Which of the model's errors keeps recurring in my trade, and I haven't written it down yet?

Where to go next

Used for a month, the hybrid decision template leaves you a map of what the model adds and what it takes away. Chapter 7 starts from that map to ask what makes you hard to replace.

Difunde el mensaje