Think First. Then Prompt.

By Jim Stroud

The test subjects were college students. Fifty-four of them, from five Boston-area schools, were wired with electrodes and asked to write essays.

One group wrote with ChatGPT. One used Google. One used nothing but their own heads. Researchers at the Massachusetts Institute of Technology (MIT) Media Lab watched their brains the whole time.

The ChatGPT group posted the weakest brain connectivity of the three. The circuits behind working memory and attention went quiet. Afterward, more than 8 in 10 ChatGPT users could not quote a single line from the essay they had turned in minutes earlier. Nearly 9 in 10 unassisted writers passed the same test.

The researchers named the effect cognitive debt. Artificial Intelligence (AI) raises your output today and charges the bill against your ability to think later.

You are the population in that study. Same age, same schools, same writing assignments. Here is what the data says to do about it.

Sequence decides everything

Buried in the same study is the finding that should change how you work. In a fourth session, students who had written alone were handed the AI. Their brains lit up across every frequency band, higher than any other group. They had already formed opinions, so the machine’s suggestions had something to push against.

Wrestling is exercise.

Think first and AI sharpens you. Prompt first and it thinks for you, and the part of you that would have done the thinking gets weaker each time.

The study was small, 54 people, and the paper has not yet been reviewed by outside scientists, so treat it as a warning rather than a law. The warning is enough.

Which one are you

A field study of 244 consultants at Boston Consulting Group (BCG), run with researchers from Harvard, MIT, Wharton, and Warwick, found three ways people work with AI. Everyone got the same tool and the same task: pick which of three brands deserved investment and write a memo to the CEO.

Centaurs, about one in seven. They ran the analysis themselves and called in AI for specific questions, like a detective checking a database. They produced the most accurate recommendations in the study.

Cyborgs, six in ten. Fused with the tool, trading ideas back and forth all day. They built real AI skills, though the researchers flagged a habit they called performative validation: going through the motions of checking the output while the judgment underneath naps.

Self-Automators, more than one in four. They handed the task to the machine and made surface edits, or none at all. Their work was the weakest, and they gained no skill of either kind. The researchers’ term for that outcome is “no skilling.”

These were trained consultants with a paycheck on the line, and a quarter of them abdicated anyway. Left alone, the drift runs toward handing it over, because handing it over feels productive right up until someone checks the work.

The paycheck follows the brain

AI skills pay. PwC’s 2026 AI Jobs Barometer, built on more than a billion job ads across 27 countries, puts the average wage premium for AI skills at 62 percent. Lightcast, analyzing 1.3 billion postings, found a more modest 28 percent, or roughly $18,000 a year.

Read the fine print and the premium tilts one way. Foote Partners found employers pay 19 to 23 percent more for demonstrated AI ability and only 9 to 11 percent for AI certifications. The market pays for people who can reason with the tool. It pays far less for people who can only operate it.

The Self-Automator gets a very confident autocomplete and a shrinking career. The Centaur gets the raise.

Why this shows up in your job search

A resume shows a recruiter none of this. A polished portfolio or a flawless take-home could be you, or the machine wearing you like a suit. That is why employers are moving to timed exercises and live case interviews, where the only brain in the room is yours.

Walk into one of those having spent four years prompting first, and the assessment will find out.

How to build the debt-free version

• Write the first draft yourself. Outline and draft before you open a chatbot. Then use AI to attack the draft. That is the order the brain data rewards.
• Play Centaur on anything that matters. Assign AI the specific queries (find, check) and keep the analysis. Save full automation for tasks where a wrong answer costs nothing.
• Read every AI output like an editor. Ask what it got wrong before asking what it got right. If you can’t find one error, you didn’t read it.
• Test yourself after. Close the tab and explain the argument out loud. If you can’t, you didn’t do the work, and an interviewer will notice before you do.
• Keep one AI-free habit. A class, a project, a weekly essay written cold. The MIT students who wrote unaided had the strongest recall and the most active brains. Keep a place where yours has to show up.

AI will write your cover letter and your take-home. It cannot sit in the interview chair. Make sure the person who does still has something running.

— Jim Stroud is a Career Intelligence Analyst, labor market strategist, and Head of Market Strategy & Industry Engagement at ProvenBase. With more than two decades of experience in recruiting, sourcing, and labor market analysis, he helps organizations and job seekers make sense of a rapidly evolving employment landscape. 

He is also the publisher of Career Intelligence Weekly (which tracks the hidden job market), author of “The Number One Job Hunting Book in the World,” and host of The Jim Stroud Podcast (commentary on the world of work). He is also an international conference speaker, job search workshop facilitator for college students and author of multiple books on career strategy and recruiting.