There’s a moment I genuinely dislike when I open my notebook to work through a problem. That moment of sitting there, staring at a blank page, asking myself: can I actually figure this out? Do I know enough? What if nothing comes?
When I open AI instead, that moment disappears entirely. AI and critical thinking don’t mix as well as most people assume.
That’s a problem.
Our brains are wired to conserve energy. Hard thinking is metabolically expensive, and given a shortcut, the brain will take it every time. This isn’t a character flaw. It’s biology. But for most of human history, the shortcut wasn’t available. You had to sit with the discomfort of not knowing, work through the fog, and arrive somewhere on your own. That friction wasn’t a bug in the thinking process. It was the thinking process.
AI eliminates the friction. You ask a question, you get an answer that sounds confident and well-structured, the anxiety of uncertainty goes away, and most people stop there.
But getting an answer is not the same as arriving at the right answer. And the gap between those two things is where a lot of smart people are quietly getting worse.
I know this because I’ve done it myself.
I recently used AI to pressure-test my pricing. It came back with numbers higher than I would have chosen on my own, and the reasoning was solid. Market benchmarks, positioning logic, the kind of structured argument that’s hard to dismiss. So I pushed my prices up. And now I’m sitting with a question I genuinely can’t answer yet: is this the growth I needed, or am I testing the edges of what the market will actually bear?
I don’t know. That answer takes time.
What I do know is that I moved forward without pressure-testing the AI’s output against people who know my market specifically. I have the experience to evaluate the logic. I skipped the step of having someone with direct context stress-test the conclusion. Not because I’m careless. Because the answer felt complete. It had reasoning behind it. It removed the discomfort of deciding alone. And finished feels like done.
That’s the trap. Not that AI gives you bad answers. It’s that AI gives you answers that feel finished. And your brain, which was already looking for a way out of the hard thinking, is very happy to agree.
Critical thinking is like physical fitness. You don’t lose it dramatically. You lose it gradually, in small surrenders you barely notice. Every time you accept an answer without examining it, every time you skip the step where you ask “but is this actually right for my specific situation,” you’re skipping a repetition at the gym. Miss enough reps over enough time and the muscle atrophies. You’re still capable of thinking hard. You just do it less, and it gets harder each time you try.
The dangerous part is that the decline doesn’t announce itself. You feel productive. You’re generating answers faster than ever. The output looks good. It’s only later, when something goes wrong or a more experienced person asks a question you can’t answer, that the gap shows up.
And it goes deeper.
Even when people realize AI led them to a wrong or incomplete answer, they rarely admit it publicly. It takes a specific kind of courage to say “I used AI to arrive at this conclusion and it turned out to be wrong.” Most people would rather obscure the process, learn quietly, and move on. Which means the feedback loop that would help others avoid the same mistake never gets completed.
This is how individual errors become collective blind spots.
The solution isn’t to stop using AI. That’s not the argument. AI is genuinely useful, and anyone not using it is leaving real leverage on the table.
The solution is to use it in the right sequence.
Wrestle with the problem yourself first. Even briefly. Even imperfectly. That initial struggle, however uncomfortable, is what gives you the context to evaluate what AI gives back. If you skip it, you have no frame of reference. You’re just accepting a well-formatted response from a system that is, at its core, an aggregate of what already exists. It can synthesize and connect and articulate. But it cannot know your business, your clients, your market, your specific risk tolerance, or the thousand contextual details that make your situation different from the average case it was trained on.
Then, after you’ve thought it through and after you’ve heard what AI has to say, talk to people who actually know the domain. Not to validate what AI told you. To interrogate it. A colleague who has navigated the same problem, an advisor who has seen it go wrong, a peer who will push back instead of agree. Human experts carry something AI cannot replicate: the memory of consequence. They’ve seen what happens when the answer that looks right turns out to be wrong in practice. That context is not in the training data.
AI as a starting point, followed by your own evaluation, followed by a conversation with someone who knows, is a defensible process. AI as a destination is just outsourcing the thinking while pretending you did the work.
I’m writing this because I want to be honest about something I’m still struggling with, not because I have it figured out. The pricing question is still open. I pushed harder than I would have without AI in the mix, and I think that’s probably good. But I also know I skipped a step I should have taken. And I think naming that honestly, including the part where I’m not sure yet how it turns out, is more useful than waiting until I have a clean lesson with a clean ending.
The one thing I’m certain of: the discomfort of not knowing is not a problem to be solved. It’s the feeling of thinking and feeling useful. And we should probably stop trying to make it go away so fast.
