In this issue:
→ Why AI-era confidence is often someone else’s thinking wearing your voice
→ How borrowed certainty fails under pressure
→ One test that separates built judgment from borrowed fluency
There’s a new kind of confidence in the workplace. It sounds informed. It presents well. It survives exactly one follow-up question.
I call it borrowed certainty: conviction assembled from outputs instead of built through reasoning.
It feels identical to earned confidence from the inside. That’s what makes it dangerous.
The research explains the mechanism. A 2025 study on cognitive offloading found AI use correlates with critical thinking at r = –0.68, and offloading itself at –0.75. The more we hand off the thinking, the less thinking capacity we keep.
But the confidence stays. Fluency remains after the reasoning is gone.
The subtle shift
Borrowed certainty shows up in tells:
→ Strong conclusions, vague reasoning
→ Fluency on the what, silence on the why
→ Certainty that collapses the moment the conversation leaves the script
The hidden risk
Leaders running on borrowed certainty make their worst calls at the worst moments: live, under pressure, off-script. Exactly when the borrowed thinking isn’t there to consult.
That’s Cognitive Atrophy’s cruelest feature. It hides until the moment you need what’s missing.
The leadership shift
The question isn’t whether you’re confident. It’s whether you could rebuild the argument from scratch if the tool disappeared.
THE RECONSTRUCTION
Fifteen minutes. Nothing open. No notes, no tabs, no tools. This one is uncomfortable by design, and the discomfort is the measurement.
Set up. Pick one position you hold confidently and have stated publicly in the last month. A strategic call, a recommendation, a view on where your market is going. It has to be something you’d defend in a room.
Close everything. Blank page. Fifteen minutes on a timer.
Rebuild the case from scratch. Write it out as if explaining to someone who disagrees:
The claim, in one sentence.
The three reasons it’s true.
For each reason, the evidence underneath it. Not “the data showed” — what data, from where, how measured.
The strongest objection, and your answer to it.
What would have to be true for you to be wrong.
Stop at fifteen minutes whether you’re finished or not.
Reading the result
You got through all five. The conviction is built. Whatever role AI played, the reasoning is yours.
You stalled at 3. You have conclusions and you have confidence. What you don’t have is the layer between them. This is the most common outcome and it’s the definition of borrowed certainty.
You stalled at 4 or 5. You’ve never seriously argued the other side, which means you’ve been agreeing with an output rather than testing one.
You couldn’t start. Useful, not embarrassing. It means the position arrived fully formed and you adopted it. Everyone has at least one of these.
The gaps are the data. Not a verdict on you — a map of which convictions are load-bearing and which are decorative. Run it on a different position next month and compare.
Two more this week
Write your view first, then consult AI. On one meaningful question. Order is everything, and it’s the single highest-leverage habit change in this entire newsletter.
Invite one hard follow-up. Say to someone whose judgment you trust: push back on my reasoning here. Discomfort is the practice.
Question of the week: Which of my current convictions could I defend without anything the tools gave me?
Before AI reshapes your organization, make sure it isn’t reshaping your ability to think.
The Human Intelligence Assessment is free and takes about ten minutes. It measures the three capabilities AI cannot replace: Awareness, Adaptability, Authenticity.
→ Take the free Human Intelligence Assessment
Lead what AI can’t.
JoshHMiller.com