Three Symptoms of the Same Cause

If you don’t know what to prompt next, the AI has already taken the wheel.

First, the story. The Atlantic just published a chilling account of what happens when technology mollifies human trust. 

My Tesla Was Driving Itself Perfectly—Until It Crashed

The danger of almost-perfect tech. Raffi Krikorian, The Atlantic 1

Douglas Adams once said, “Technology is a word that describes something that doesn’t work yet.” Hold onto that until we stop calling AI a technology. This ‘almost-perfect tech’ is running loose in our writing, our systems, our outputs, and, it turns out, our thinking.

Two Papers, One Diagnosis

Two research papers deepen this story. Both point to our use of AI as the culprit.

AI makes you smarter but none the wiser: The disconnect between performance and metacognition. Computers in Human Behavior Volume 175, February 2026, 108779

No one crashes in this paper, but the impacts hit just as hard. Our AI usage improves what we produce, but does nothing for us. The paper is worth scanning through. And if you think your AI Literacy is high, Section 6.1 is written for you.

Here’s the number that lands: 46% of participants prompted the AI only once per question. Only 8% went past three prompts. That is using AI as a vending machine. Punch the keys, walk away with a tasty snack. Good in an emergency, no nutrition, no sustenance.

The Variance Paradox: How AI Reduces Diversity but Increases Novelty. Bijean Ghafouri, University of Southern California (2025).

This paper shows what happens when AI puts your original thought through a crash test simulator. What comes out fits the AI’s narrow sense of what your idea should be. Smoothed. Compressed. Safer. But the Variance Paradox also points to a way through: we can recover what we’re losing by changing how we interact with AI. Push for counter-narratives. Interrogate your own results. Curate and spar with the AI. This is how you sharpen your arguments and your thinking. 

Both papers point to the same gap. We don’t do the one thing that works: prompt again. We are far better off using AI as a sparring partner trying to destroy our intention than as an affable golden retriever ready to fetch whatever we throw.

Your ‘Almost-Perfect’ Driving Test

If you don’t know what to prompt next, your hands may be on the wheel, but you are not driving. That’s the clearest sign you’ve become a passenger. Here are three follow-up prompts that help you wrestle back control:

1. Ask what phrases in this response could apply to any company, any argument, any topic.

2. Ask what decisions were made that were not specified.

3. Ask where the awkward phrasing was improved.

Oh, and ask the AI to detect, not fix. Diagnose first. You write the corrections.

1‍ ‍Yes, it is behind a paywall. Human-written journalism costs more than AI. It is worth it. 12 mos Atlantics = 4 mos Claudes, and it will likely be the best writing you are exposed to for the next year. As a bonus, it’s also a masterclass in intentional storytelling.