Is AI Skepticism Really Wrong, the Productivity Myth Meeting Its Empirical Void

There is a strange symmetry on Hacker News these days. One post, defending AI coding tools against lukewarm colleagues, declares that the skeptics are simply out of their minds. Another, posted within the same window, confesses that large language models are dismantling the author's professional foundation. Both drew thousands of comments. Both were, in their way, certain. One side mocks those who refuse the tools as fools; the other fears the tools will render its own skills obsolete. What unites them is not the conclusion but the confidence. Measured by conviction alone, the evangelist and the doomer are not enemies but twins raised in the same epistemic soil, a soil in which feeling has quietly been promoted to the rank of evidence.

Faith and dread stand on the same empty ground

The productivity gospel travels by anecdote. A weekend project finished in an afternoon, boilerplate that no longer needs typing, a vague but vivid sense of acceleration. These reports may be sincere, but sincerity is not measurement. Between the fact that one engineer finished one task faster and the claim that engineering as a discipline has become more productive lies an enormous, unexamined leap. Generating code quickly is not the same as completing a maintainable system quickly, and the latter is what the craft actually consists of. The pessimist's case is no firmer. The dread of career erosion draws not on employment data or studies of how job roles are being restructured, but on an emotional extrapolation from impressive demos to an imagined personal future. Each camp holds its certainty tightly, yet neither possesses the controlled measurement that would justify it.

The few well-designed randomized studies in this space are unsettling precisely because they refuse to flatter either side. One widely discussed experiment found that experienced open-source developers using AI assistance felt faster while their measured completion times actually grew longer. This is not a universal verdict, and it should not be wielded as one. Its real lesson is narrower and more disquieting: perceived speed and measured speed can diverge sharply, and almost every claim that animates this debate lives on the perceived side of that gap. The word productivity has become a slogan for both armies without anyone agreeing on its unit.

Inevitability is crowding out verification

What distorts the conversation further is the rhetoric of inevitability, the insistence that the technology is already flowing in a settled direction and only the maladapted will be swept aside. This framing transforms the basic empirical question into idle hand-wringing. Curiously, inevitability serves both camps at once. For the evangelist it supplies momentum, a reason to adopt now and ask questions later. For the pessimist it supplies an alibi, a way to rationalize a sense of powerlessness as simple realism. The casualty in both cases is the same: the question of exactly how much this tool helps or harms on exactly which tasks gets pushed off the table by people too busy proclaiming or mourning to measure.

In a healthier discourse, skepticism would not be the enemy of the tool but the method for assessing it honestly. Good skepticism does not reject the technology. It demands that anecdote be converted into measurement, demos into reproducible experiments, impressions into data. What the developer community needs now is not greater conviction but a more disciplined admission of ignorance. We do not yet know what AI coding tools change, and the real expertise of this moment lies in the patience to fill that ignorance with measurement rather than with faith or fear. The louder the polarized shouting grows, the more valuable the quiet work of accumulating evidence becomes.