The Moment You Nod Along and Move On
There’s a particular moment I’ve started noticing in my own work, and I’d bet you’ve had it too. You ask AI for something, it hands back a response that’s clean, organized, confident, and reads like it knows exactly what it’s talking about … and you nod along and move on. There’s no friction and no second look. The thing sounded finished, so you treated it as finished.
The trouble is that “sounds finished” and “is actually right” turn out to be two very different things, and the gap between them is precisely where your judgment should live.
I came across a piece recently by Rashi Desai in Towards Data Science that put a name to the thing I’d been circling. She argues that the people getting the most out of AI aren’t the best prompters, they’re the ones regulating their own thinking while they use it. The skill has a slightly clunky academic name, metacognition, which is really just a tidy way of saying thinking about your thinking. (Cognitive scientists have been studying it for decades. The rest of us are suddenly going to need it.) Her prediction is the line that stuck with me, that “prompt quality will be commoditized but cognitive discipline won’t be.”
If you’ve heard me talk about AI raising the floor, you can probably see where this is heading.
When Everyone Has the Same Tool
According to McKinsey’s State of AI 2025, 88% of organizations now use AI in at least one business function, up from 78% a year earlier. In other words, the tool stopped being the advantage a while ago. Everyone has the same tool, the same speed, the same eerily fluent outputs. (How novel.) What’s left to set you apart is the part the machine can’t do for you, which is the thinking you bring to it and the thinking you do with it.
What Happens When the Output Looks This Good
The part that nags at me is what happens to our thinking when the outputs get this good.
Researchers at Microsoft and Carnegie Mellon surveyed 319 knowledge workers about critical thinking and AI, and found something that should make all of us sit up straight. The more confidence people had in the AI, the less critical thinking they did. The more confidence they had in their own expertise, the more they questioned what the AI handed them. (Read that twice. Trusting the tool more made people think less.)
The same study turned up a detail that lands right on top of the ordinariness problem. People with AI access produced “a less diverse set of outcomes” for the same task than people without it. Everyone drifted toward the same answers, which is more or less the definition of becoming ordinary, just dressed up in research language.
Then there’s the MIT Media Lab study with the wonderfully ominous title “Your Brain on ChatGPT,” where researchers wired people up with EEG sensors while they wrote essays. The group leaning hardest on AI showed the least neural engagement, remembered the least about what they’d written, and felt the least ownership over the result. The researchers called it “cognitive debt,” which is one of those phrases I wish I’d coined. (Borrowing mental effort from your future self, at interest. … sigh.) Worth saying these are early studies rather than the final word, but the direction of travel is hard to wave away.
The more confidence we put in the tool, the less thinking we do. That’s a discipline problem, and discipline can be trained.
How to Think About Your Thinking
So what does it actually look like to think about your thinking while you work with AI? Less complicated than the academic language makes it sound, as it turns out, and it mostly comes down to staying mentally present instead of slipping into autopilot. A few things I try to do, with genuinely varying success on a busy day.
- Challenge the output before you accept it. Argue with it. Ask what it’s assuming, where it might be leading you astray, what a sharp critic would tear into first.
- Sit with the uncomfortable bit. AI is brilliant at making confusion disappear, but your most original thinking tends to live inside that confusion, so resist letting it get resolved away too quickly.
- Notice when you’re agreeing because it sounds smart. Fluent and correct aren’t the same thing (see above, repeatedly).
- Use AI to find your blind spots, not to confirm what you already believe. Ask “what am I not seeing” far more often than you ask it to tell you you’re right.
- Keep ownership of the judgment. Let it draft, analyze, and stress-test all day long, the decision about what to do with any of it stays with you. (This is the Delegation Dial question in miniature, knowing which moments are AI’s and which ones are yours.)
A small shift in how you ask makes a surprising difference. Instead of “summarize this and give me recommendations,” in my experience you get something far more useful from “summarize this, tell me what you’re assuming, where the data might mislead me, and which conclusions wouldn’t be justified.” Same tool. Completely different relationship with your own brain.
Where the Thinking Actually Lives
If you’ve sat through one of my talks, you know I’m fond of the 20-60-20 Collaboration Equation, where you invest the first 20% in framing and context and judgment about what’s even worth doing, AI handles the middle 60% of generating and drafting and synthesizing, and you own the last 20% of review, validation, and making the thing genuinely yours. I’ve always taught the bookend 20s as the human part. What the metacognition research clarifies for me is that both of those 20s are thinking work, and thinking of a very particular kind. The first 20 is the discipline to frame the problem properly before you reach for the tool. The last 20 is the discipline to interrogate what comes back rather than waving it through. The 60 in the middle is where the speed lives. The 20s on either end are where you stay a person worth hiring.
Speed is the easy part to buy now. Discernment is the part you still have to bring yourself.
The floor is going to keep rising, and the outputs are only going to get more fluent and more convincing, which makes the temptation to switch off your own thinking stronger, not weaker. The reassuring part is that the antidote isn’t technical and doesn’t require a course. It mostly asks you to stay awake at the wheel and keep checking whether the thing in front of you is actually good or merely well-formatted, which is another way of saying it asks you to treat your own judgment as the asset it became while everyone else was busy admiring the speed.
AI can generate almost anything now. What it can’t do is care whether the answer is right. That part, thankfully, is still yours.
• • •
One last thing before you go. Not many people realize this, but I run programs specifically on critical thinking with AI, which means the master’s in communication is finally earning its keep (somewhere, the professors who taught it are delighted). I get to pull on all my innovation research and inventor habits too, and the sessions have picked up some genuinely lovely reviews, mostly because they give people a different, practical, and positive way to build real AI empowerment rather than another doom-laden lecture about the robots coming for their jobs. If that sounds like something your team could use, you know where to find me.
Frequently Asked Questions
What is metacognition when it comes to using AI?
Metacognition means thinking about your own thinking. In an AI context, it’s the awareness that lets you notice when you’re accepting an answer because it sounds convincing, when you’re being intellectually lazy, or when the tool is narrowing your reasoning instead of expanding it.
It’s an internal monitoring system, and it becomes essential the moment AI starts producing work that feels complete even when it’s shallow or subtly wrong. The fluency of the output is exactly what makes it easy to stop paying attention.
Why does this skill matter more as AI gets better?
As AI improves, the quality and speed of its outputs stop being a differentiator, because almost everyone has access to the same capability. What sets people apart is the thinking they bring to the tool and the discipline they apply to what it gives back.
McKinsey reports that 88% of organizations now use AI in at least one function, so the tool itself is table stakes. When everyone can generate, the advantage moves to whoever can still think independently about what’s worth generating and whether the result holds up.
Does using AI actually reduce critical thinking?
Research suggests it can, particularly when people trust the AI more than their own expertise. A Microsoft and Carnegie Mellon study of 319 knowledge workers found that higher confidence in AI was associated with less critical thinking, while higher confidence in one’s own skills was associated with more.
A separate MIT Media Lab study found that people who relied heavily on AI to write showed lower cognitive engagement and felt less ownership over their work, an effect the researchers labeled cognitive debt. These are early findings rather than settled science, but the pattern is consistent enough to take seriously.
How can I keep thinking critically while using AI?
Stay mentally present rather than slipping into autopilot. Challenge the output before accepting it, ask the AI what it’s assuming and where it might mislead you, and use it to surface your blind spots rather than confirm what you already believe.
The practical version is to change how you ask. Rather than requesting a summary and recommendations, ask the tool to flag its assumptions, identify where the data could mislead you, and name the conclusions that wouldn’t be justified. You keep the judgment, the AI does the heavy lifting.
What’s the difference between an AI user and an AI thinker?
An AI user tends to outsource thinking in exchange for speed, accepting outputs at face value. An AI thinker uses the same tool to stress-test, expand, organize, and challenge their own reasoning, treating it as a partner rather than a replacement.
The distinction shows up in the questions people ask. Instead of “give me the answer,” the stronger users ask what assumptions they’re missing, what would invalidate their argument, and what perspective they’ve ignored.
How does the 20-60-20 Collaboration Equation fit into this?
The 20-60-20 model puts humans in charge of the first 20% (framing and context) and the last 20% (review and personalization), with AI handling the middle 60%. Both human bookends are thinking work, which is exactly where metacognitive discipline applies.
The first 20 is about framing the problem well before reaching for the tool, and the last 20 is about interrogating the result instead of accepting it. The middle is where the speed comes from, but the value lives in how well you handle the two ends.
Is this a leadership skill?
Increasingly, yes. In organizations with heavy AI adoption, the bottleneck is no longer access to information, it’s discernment. Leaders face cognitive overload and an abundance of confident-sounding inputs, and the ones who can regulate their own thinking will make better sense of it.
The leader’s role shifts from having the answers to being able to evaluate an overwhelming volume of inputs without defaulting to whatever the AI produced. Modeling that discipline for a team may turn out to be one of the more valuable things a leader can do.
Will prompting skills still matter?
Prompting still matters as a foundational skill, but it’s likely to become commoditized as tools get better at interpreting intent. The harder-to-replace skill is cognitive discipline, the ability to stay intentional and independent in your thinking while you work.
Prompt quality can be taught, copied, and eventually automated. The discipline to question, refine, and own your own judgment is far more personal, which is what makes it durable as the technology keeps improving.
