The Comfort Plateau AI Built For You
I’ve spent the last year watching engineers and writers adopt these tools, and the same pattern keeps showing up. The first three months look transformative. The next nine look almost identical to the first three.
The output keeps coming. The confidence keeps building. The skill stops moving.
This isn’t a story about AI making people dumber. It’s a story about a comfort plateau that didn’t exist before, and what it costs to stay on it.
The accessibility breakthrough is real
Draw a spider chart. Hundreds of axes: SQL, contract law, React, radiology, negotiation, Kubernetes, copywriting. Score a human. Most experts look like a star with two or three spikes and a flat field everywhere else. Depth is the point.
Plot AI on the same chart. It’s not a 5 on any axis. But it sits at a 2–4 across nearly everything. No individual has ever had a tool this broadly competent.
This is why the gains for newcomers are so large. A BCG study of 758 consultants found below-average performers improved by 43% and above-average performers by 17%. A field study of 5,179 customer-support agents found novices gained 34% in productivity while the most experienced gained almost nothing. Going from “can’t do this” to “shipped a thing” used to take months. Now it takes minutes.
This part of the story is good. It’s also where most writing about AI stops.
The bicycle you can’t draw
Ask a person to rate their understanding of how a bicycle works. They will rate it high. Ask them to draw one from memory. About half of non-cyclists get the chain wrong, looping it around both wheels instead of only the back wheel and pedals. Many draw the frame joining the front and back wheels, which would make steering impossible. Asked to re-rate after the drawing, confidence collapses.
This is the Illusion of Explanatory Depth. People feel they understand things they can’t actually explain. The illusion only breaks when you’re forced to produce the explanation. Without that production step, confidence sits unchallenged forever.
AI removes the production step.
You don’t draw the bicycle. You ask for one. The pedals are in the right place. The chain runs correctly. You ship it. Nothing collapses. The next time you need a bicycle, you ask again, and the feedback loop that would have built a mental model never closes.
I’m not certain the bicycle effect maps perfectly onto code or prose. But the mechanism, fluent output without forced explanation, feels right. And there’s now direct evidence: an Anthropic study measured exactly this in January. Junior developers learning an unfamiliar library with AI assistance scored 17% lower on conceptual understanding and debugging than those who learned without it. The code they shipped looked indistinguishable. The judgment behind it did not develop.
“But we had this fear before”
Calculators were supposed to kill arithmetic. Spell-check was going to end literacy. Stack Overflow would breed shallow programmers. Most of those predictions were wrong. A meta-analysis found calculator use, integrated properly, improved math skills. Stack Overflow made code more functional but significantly less secure, and yet didn’t atrophy programmers broadly. Why would AI be different?
Because of where it removes the work.
A calculator removes execution. You still set up the problem. The mental model of “I need to multiply these two things and check the units” stays yours. Stack Overflow removes retrieval. You still adapt the answer, integrate it, decide whether it fits.
GPS removes generation, the building of the spatial map. A longitudinal study tracking GPS users over three years found that heavy users showed declining hippocampal-dependent spatial memory. Not because they were already bad navigators. Because they stopped generating maps.
AI is like GPS, not like a calculator. It removes the generation step, the building of the mental model. That’s the step that triggers the bicycle collapse. Take it away and the collapse never happens.
The motivated minority objection
There’s a real counter-argument: every tool produces a distribution. The top quartile uses it to accelerate; the median plateaus; nothing changes about who pulls ahead. AI is just the latest example.
This is half right. The thing that used to make some people more valuable than others, deep knowledge in their specific domain, matters less now. AI has narrowed that gap on the work it can do. The BCG study showed it explicitly: bottom-half consultants gained more than top-half. The expertise gap is closing on AI-shaped tasks.
But a new axis is opening. Variance among AI users isn’t shrinking; it’s growing by about 47% in some measures. The new top quartile isn’t the people with the deepest domain knowledge. It’s the people who can sit with an AI output and ask: what would I have to draw to know if this is right? The ones who treat fluency as a warning, not a signal.
That’s a skill the AI cannot give you. By design, it removes the very moment that develops it.
The plateau is where the squeeze is
There’s an external reason this matters now, not just a personal one. The roles being compressed aren’t the ones at the top of any skill chart. They’re the ones whose required skill sits at or below what AI already provides. AI raises the floor. Roles at or under the floor get cheaper, fewer, or absorbed into someone else’s job. Roles above the floor, where the work requires understanding deep enough to explain and not just outputs polished enough to ship, don’t.
The comfort plateau is the floor. Stay there and the work you’re doing is work AI already does. The work AI cannot do for you is the work that will keep belonging to humans.
What this asks of you
Try this. Take something you shipped with AI last week. Explain how it works, out loud, without looking. Notice where the explanation gets thin.
If most of it gets thin, that’s the plateau. Not a failure of effort. A failure of friction.
Five years from now, the largest skill gap in knowledge work won’t be between people who use AI and people who don’t. It will be between people who let the fluency pass unchallenged and people who insisted on drawing the bicycle themselves.
The tool will keep working either way. The growth will not.