The dependency arrived faster than anyone predicted. In surveys and workplace anecdotes now accumulating across the tech industry, a striking pattern has emerged: a meaningful share of professional programmers say they will not accept jobs—or will leave current ones—if denied access to AI coding assistants like GitHub Copilot, Cursor, or Claude. What began as a productivity enhancement has become, for many, a psychological crutch they cannot code without.
This is not, on its face, irrational. AI pair-programming tools genuinely accelerate boilerplate generation, catch syntax errors in real time, and reduce the cognitive load of context-switching between documentation tabs. Developers who use them report feeling more productive and less frustrated. The tools work. The question is what happens to the humans who use them.
The atrophy hypothesis
Critics within the programming community have begun articulating a concern that sounds almost reactionary but carries empirical weight: that relying on AI to generate code erodes the deep, deliberate practice that produces genuine expertise. Junior developers who never struggle through a recursive algorithm unaided may never internalize why it works. Senior engineers who outsource debugging to a model may lose the intuition that once made them senior.
The parallel to GPS navigation is instructive. Studies have shown that heavy GPS use correlates with reduced spatial memory and hippocampal gray matter. The tool works; the user's underlying capability declines. Programming is more cognitively complex than driving, but the mechanism—offloading effort to an external system—is identical.
The leverage problem
There is also a labor-market dimension. Developers who insist on AI access as a job requirement are, in effect, telling employers that their productivity is contingent on a third-party subscription. This is not a strong bargaining position. If a programmer's output collapses without Copilot, the employer's rational response is to value the tool more and the programmer less. The dependency that feels like empowerment is actually a transfer of leverage from worker to platform.
Meanwhile, companies are watching. The same surveys that reveal developer attachment to AI tools also inform corporate calculations about headcount. If one AI-assisted engineer can do the work of two unassisted ones, the second engineer is redundant. The workers most vocally dependent on AI may be accelerating their own obsolescence.
Our take
The refusal to work without AI is a category error dressed as a preference. Chainsaws are better than axes; no one demands the right to chop wood by hand. But programming is not wood-chopping. It is a craft whose value lies partly in the practitioner's irreplaceable judgment—judgment that atrophies when the hard parts are outsourced. Developers who cannot function without AI are not more productive; they are more fragile. The smart play is to use the tools heavily and retain the ability to work without them. That combination is rare, and rare skills command premiums. Dependency commands nothing.
