Why AI makes you feel like you understand more than you do — and what that costs you long-term
You shipped the feature. The tests pass. You can explain what the code does.
But ask yourself: could you have built it without the AI?
Not "would it have taken longer?" — could you have done it at all, from scratch, on a desert island with no internet?
If that question makes you uncomfortable, you're not alone. And it's not imposter syndrome.
It's something worse: the calibration gap.
Calibration is the match between your confidence and your actual ability. Good calibration means you know what you know, and you know what you don't. Poor calibration means your confidence and your competence have drifted apart.
AI tools are exceptional at creating calibration gaps. Here's why:
When you ask an AI a question and get a correct answer, your confidence goes up. You now "know" the answer. But the mechanism that increased your confidence was the AI's reasoning, not your own. Your actual understanding didn't change — only your belief about your understanding did.
This is different from reading a tutorial. When you read a tutorial, you follow the reasoning step by step. You build a mental model. Your brain does the work.
When you ask an AI and it answers, your brain can skip the work. The answer arrives. The curiosity satisfied. The feeling of learning without the learning.
When you learn something the hard way — through struggle, failure, debugging, iteration — your brain encodes it deeply. The difficulty of retrieval creates the permanence of memory. This is what Robert Bjork calls desirable difficulties.
AI removes the difficulty. And in doing so, it removes the encoding.
You get the answer. You understand it at that moment. You move on. Two weeks later, the answer is gone. But your confidence from that moment of understanding? That stayed.
Calibration gaps hurt senior engineers most, and here's why:
A junior engineer who uses AI for everything will quickly notice they're lost. The gaps are so large they become obvious. They know they don't know.
A senior engineer with an AI assistant? The gaps are subtle. They can still reason. They can still debug. They can still ship. But the deep, embodied understanding — the kind that takes years to build and shows up at 11pm when something breaks in production — that's eroding quietly.
Senior engineers are confident and skilled, so when AI inflates their confidence further, they don't notice the skill drift. The math looks like this:
Actual skill: 85/100 AI-boosted confidence: 92/100 Calibration gap: 7 points
That 7-point gap is invisible. It shows up as:
You're still a good engineer. But you're not as good as you think you are. And that gap is growing every sprint.
"I don't deserve this success; I'm fooling everyone."
Makes you anxious. Visible. Pushes you to over-prepare and prove yourself.
"I deserve this success; I understand this."
Makes you confident and wrong. Invisible. Pushes you to move faster and trust outsourced instincts.
Imposter syndrome is visible. The calibration gap is not.
If someone asked you to implement something fundamental to your stack — a caching layer, an authentication flow, a data pipeline — could you do it without looking anything up? Not "would take a while" — could you do it at all, right now, cold?
When you ask an AI a question and it answers, does the answer surprise you? Do you learn something new? Or does it mostly feel like hearing your own thoughts articulated back? Familiarity is a signal your brain isn't processing new information — it's recognizing patterns it already has.
Do you regularly underestimate how long things take — or do you overestimate how much you can accomplish in a sprint? Calibration gap makes you optimistically wrong because you genuinely believe you understand the work better than you do.
The fix isn't "use less AI." That's treating a symptom, not the mechanism.
The fix is deliberate calibration:
The calibration gap won't make you a bad engineer. Not immediately. Not even perceptibly.
It will make you slightly less sharp each month. Slightly more dependent on the AI for confidence. Slightly less able to operate without it.
And one day — probably at the worst moment — you'll notice. A question you can't answer. A bug you can't find. A design you can't defend.
That moment won't feel like a crisis. It'll feel like Tuesday.
If this landed, read The Science of AI Fatigue — the research behind the calibration problem, including Bjork's desirable difficulties, Bainbridge's ironies of automation, and what the neuroscience actually says about how we learn (and what destroys learning).
Take the AI Fatigue Quiz and see where you stand.
Take the Quiz →