Where Does a Leader’s Judgment Belong When AI Writes the First Draft?
The AI draft of the board update reads beautifully. The numbers line up. The tone is right. A leader reads it once and thinks: this is obviously good.
When AI writes the first draft, a leader’s judgment has to move to one job: choosing which part gets real scrutiny. Feeling confident is not a reliable sign the draft is right.
I’m Sahar Andrade, MB.BCh, Neuroleadership Coach and Forbes Coaches Council member. In my experience, steady leaders do not check everything. They decide what to check.
How has AI changed what leaders do with a draft?
Now the first draft often comes from AI. In a 2026 Wharton Human-AI Research and GBK Collective survey of more than 1,400 U.S. respondents, reported by Fortune on October 13, 72% of managers and above said their focus has shifted from creating outputs to managing AI-generated outputs. That was 63% of managers and directors, and 82% of vice presidents and above.
Many leaders hear that and think the job got lighter. Reviewing feels easier than creating.
My read is different. Part of the leader’s job is deciding what needs checking. When a person writes a draft, the thinking lives inside the writing. When AI writes it, the human thinking has to happen somewhere else. For leaders, I believe that somewhere else is the decision about what to check.
Why can a leader feel sure about an AI draft that is wrong?
In January 2026, Wharton researchers Shaw and Nave posted a working paper called “Thinking, Fast, Slow, and Artificial.” It is a preprint, not yet peer reviewed.
They ran three preregistered experiments with 1,372 participants and 9,593 trials, using an adapted Cognitive Reflection Test: short reasoning puzzles where the first answer is often wrong. People could consult an AI. The researchers secretly varied whether the AI was right.
People consulted the AI on more than half of trials. In Study 1, compared with no AI, accuracy rose 25 percentage points when the AI was right and fell 15 points when it was wrong.
Here is the finding that matters most. Using the AI raised confidence, even after errors. People felt more sure whether the answer was right or wrong. In these experiments, confidence was a poor signal of accuracy when the AI was wrong.
Time pressure did not remove the pull of the AI’s answer on accuracy. Incentives and feedback reduced it, but did not remove it. People with higher trust in AI showed more of it. The authors call adopting AI output with minimal scrutiny “cognitive surrender.”
What does “cognitive surrender” mean for executives?
“Cognitive surrender” is a proposed research construct, a name for a behavior seen in the data. It is not a clinical label for anyone. It is not proof that executives have lost independent judgment.
These studies tested reasoning puzzles, not executive leadership. Applying them to leaders is my inference, not their finding.
A draft can read smoothly and still contain errors. If the feeling of “this is right” shows up either way, a leader cannot use that feeling to choose what to check. The choice has to come from consequence instead.
How should a leader check an AI draft under pressure?
What follows is a practical recommendation informed by this research. It is not an intervention these experiments tested. The studies describe the problem. This is my response as a coach.
Name the costly part first. Before reading the draft, name the one part where an error would cost the most. Four places to start: a number, a promise, the people affected, or a legal claim.
Check it against the source. Rereading the draft is not checking it. Open the finance file, the contract, the original email. If the source is a person, ask them.
Read confidence as a feeling, not a fact. If the draft feels obviously right, that is the moment to check the costly part anyway.
Under time pressure, check what matters before signing off. Start with the costly part, then anything else that could change the decision. Compare the claim with its original source. If there is not enough time, delay the sign-off or get help.
Picture a customer email drafted by AI. The costly part is the delivery date. A quick look at the project schedule shows “end of month,” not “next week.” One sentence changes. A broken promise never happens.
If a leader has to check it, what is the AI for?
When the AI was right, accuracy rose 25 points. The tool helped a lot. The point is not to check everything. That erases the time saved, and few leaders can sustain it.
The point is to start with the part that matters most. The AI carries most of the writing. The leader carries the consequence. Those are different jobs, and the second one stays with the leader.
Frequently Asked Questions
Where does a leader’s judgment belong when AI writes the first draft?
It belongs in choosing what gets real scrutiny. When AI produces the first version of a memo, update, or email, the leader’s most valuable move is deciding which part would cost the most if it were wrong, then checking that part first against the original source, and then anything else that could change the decision. In my experience, this is more realistic than reviewing every line and more reliable than trusting a feeling of confidence. The AI can carry most of the writing. The leader stays responsible for deciding where the truth must be confirmed.
What is cognitive surrender?
Cognitive surrender is the term Wharton researchers Shaw and Nave use for adopting AI output with minimal scrutiny. It comes from a January 2026 working paper that has not yet been peer reviewed. It is a proposed research construct, a way to name a behavior seen in their experiments, not a clinical label for anyone. It also does not prove that executives have lost independent judgment. The studies used reasoning puzzles, so applying the idea to leaders is an inference.
Why is confidence a poor guide to whether an AI draft is right?
In the Shaw and Nave experiments, using the AI raised people’s confidence even after errors. They felt more sure whether the AI had led them to the right answer or the wrong one. That means, in those studies, confidence did not reliably flag when the AI was wrong. The research used reasoning tasks, not leadership decisions, so applying it to executives is my inference. Still, it suggests a leader should not use the feeling of certainty to decide what to skip.
Which parts of an AI draft should a leader check first?
Start with the part where an error would cost the most. Four places to start: a number, a promise, the people affected, or a legal claim. A board update often hinges on a figure. A customer email hinges on a date or commitment. A reorganization note hinges on who is named and how. Name that part before reading the draft and check it against the source rather than against the draft itself. Then check anything else that could change the decision.
Does time pressure remove the risk of AI errors slipping through?
No. In the research, adding time pressure did not remove the pull of the AI’s answer on accuracy. That study did not test executives, so this is an inference. My read is that busy days are when a fluent draft is most tempting to accept as written. The practical answer is to check what matters before signing off: the costly part first, then anything else that could change the decision. If there is not enough time, delay the sign-off or get help.
Should leaders trust AI drafts less?
Not necessarily. In the same research, accuracy rose 25 percentage points when the AI was right, which is a large gain. The aim is not distrust. It is placement. Interestingly, people with higher trust in AI showed more of the effect. In my opinion, that makes a fixed check worth more, not less, for leaders who rely on AI. Deciding in advance which part to verify first means the check does not depend on how sure anyone feels in the moment. The leader keeps the tool and keeps the judgment.
Want to think through where your judgment needs to sit as AI takes on more of the first draft?
Book a Leadership Clarity Call. 30 minutes. No pitch. Just clarity. https://calendly.com/saharandrade
About the author: Sahar Andrade, MB.BCh, is a non-practicing physician turned Neuroleadership Coach and a Forbes Coaches Council member. She teaches executives how the nervous system shapes leadership under pressure, and hosts the AI Café Conversations podcast.
Comments