AI Prompt Hackers

AI Prompt Hackers

AI in Action

Are we outsourcing thinking to AI or finally thinking together?

What "co-cognition" with AI might really look like.

Apr 01, 2026
∙ Paid

Here’s something that millions of people do every single day. They open an AI chatbot, type something in, read whatever comes out, and then carry on with their lives as though they definitely still have a functioning brain. The whole transaction takes about nine seconds. It has the cognitive footprint of pressing a lift button. And then, … they put that output into a report, or an email, or a strategy document, and sign their name to it.

Researchers at Microsoft and Carnegie Mellon University surveyed 319 of these people. They gathered 936 real-world examples of knowledge workers using generative AI at work. What they found, and I want you to sit with this for a moment, is that in roughly 40% of tasks, the workers applied no critical thinking whatsoever.

Zero. They read the machine’s answer and went: yes, that’ll do.

The researchers noted that mechanising routine tasks deprives users of the routine practice needed to keep their judgement sharp, leaving it “atrophied and unprepared” when anything difficult actually comes along.

Atrophied. That’s the word a Microsoft researcher chose. About people who work in offices. With computers. In 2025.


Natural born cyborgs

Before we get to the full horror of what this means, there’s a philosophical argument that keeps turning up in the academic literature on this subject, and it’s worth explaining because it reframes the whole thing in a way that is either reassuring or deeply not reassuring, depending on your disposition.

In 1998, Andy Clark and David Chalmers published a paper called “The Extended Mind.” Their point was that the boundary between what’s inside your head and what’s outside it is a lot blurrier than we like to think. They used the example of a person with Alzheimer’s who writes appointments in a notebook and checks it reliably. That notebook, in some meaningful sense, is doing some of the thinking. It’s not just a tool. Part of the cognitive process genuinely lives out there, on the page.

Clark has spent the decades since developing this idea, and in May 2025 he applied it directly to AI in a paper in Nature Communications. His argument’s that humans have always built hybrid thinking systems, ones that pull in non-biological resources, and that AI is just the latest and most powerful version of this. We are, he says, “natural-born cyborgs.”

I find this framing simultaneously correct and extremely easy to weaponise by people who want an intellectual-sounding reason to never question a chatbot again.


What happens when you build the hybrid badly

The Microsoft study is a detailed portrait of exactly that. The more confident a worker was that the AI could handle a task, the less thinking they did. The outputs got more homogenous across users. Everyone started producing subtly similar work because they were all drawing from the same machine and none of them were pushing back on it. The individual voice, the inconvenient angle, the thought that only you would have had: gone. Smoothed away by a system specifically designed to give you something fluent and acceptable, which isn’t the same thing as something true.

Research published in Frontiers in Artificial Intelligence gives this pattern a name. It’s called the “hollowed mind.” The idea is that when AI-generated answers are available instantly and effortlessly, users start bypassing the effortful thinking that actually builds understanding. The same paper identifies something it calls the “sovereignty trap,” which is where the AI’s air of confident authority nudges users into handing over their intellectual judgement, mistaking the ability to retrieve information for the ability to think about it.

The hollowed mind doesn’t arrive in one catastrophic moment. It accumulates across dozens of small surrenders: accepting a framing you didn’t choose, skipping the source because the summary seemed fine, letting the machine make a call you were perfectly capable of making yourself. I’m not pointing fingers here. I’ve done all of those things. I did some of them this morning.


The bit where it gets more complicated

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2026 Andy Wood · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture