Consider two programmers at a hedge fund. The first maintains the trading engine—a million lines of C++ that executes billions in transactions daily. The second writes scripts to parse quarterly earnings reports for the research team. Both write code. Both are software engineers. But the distinction matters because we’re now flooding both programmers with the same AI tools, expecting similar productivity gains, and wondering why the results diverge so dramatically.
<aside> 👉
The truth is, we’ve been treating “code” as a monolithic substance when it’s actually a spectrum of activities with fundamentally different physics, economics, and consequences.
</aside>
Before we can have an honest conversation about AI’s impact on software development, we need to acknowledge this hierarchy. Not all code is created equal—and pretending otherwise is why our predictions about AI coding keep missing the mark. There’s some frothy stuff happening in vibe-coding more on that later.
Charles Nahl’s “Sunday Morning in the Mines” (1872)
In 1850, California’s gold rush democratized prospecting. What followed wasn’t ten times more gold extracted—but ten times more holes dug. The parallel to our AI coding moment is instructive. The initial data seems paradoxical: despite 84% of developers using AI assistants and reporting massive productivity gains [1], we haven’t witnessed the expected Cambrian explosion of traditional software. Domain registrations, a proxy for new public-facing websites, crept up a mere 1.2% in 2024 [2].
This is our Shovelware Paradox: if developers are suddenly 10× more productive, where are the 10× more applications?
The answer lies not in a failure of productivity, but in a classic economic phenomenon: the Jevons Paradox. First observed in 1865, the paradox states that as technological improvements increase the efficiency with which a resource is used, the total consumption of that resource does not decrease—it increases, as the lower cost of use opens up new applications and drives higher demand [3]. We made steam engines more efficient, and we burned more coal, not less. We are making software development more efficient, and we are creating more software, not less.
But, the explosion isn’t happening in the world of venture-backed startups registering new .com domains. It’s happening on deployment platforms like Vercel and Netlify, which are becoming the proving grounds for this new wave of AI-generated applications.
A comparative analysis of Vercel and Netlify, two of the largest deployment platforms, reveals the true story. While both platforms started at a similar scale in early 2022, their growth trajectories have diverged dramatically, correlating strongly with the release and integration of powerful AI coding tools.
 Source: BuiltWith, Nov 2025 [4] [5]
Vercel’s growth didn’t just happen; it accelerated in lockstep with the AI revolution. A significant inflection point occurred after the release of Anthropic’s Claude 3.5 Sonnet in June 2024—a model praised for its coding capabilities—and Vercel’s own major v0 generative UI launch in October 2024 [6] [7]. In the last three months of 2025 alone, Vercel added over half a million new sites, 2.6 times more than Netlify.
This isn’t just market share redistribution; it’s market expansion. The 1.87 million net new sites on these platforms represent applications that likely would not have existed otherwise. The efficiency gains from AI haven’t just made existing developers faster; they have lowered the barrier to entry so dramatically that a new class of creators is now shipping code.
This is the Jevons Paradox in action. The “resource” is software development, and the “efficiency gain” is the widespread adoption of AI coding tools like GitHub Copilot, which has grown to over 20 million users, and Cursor, which has seen explosive revenue growth [8]. The result is not less work for developers, but an exponential increase in the amount of software being created, particularly the disposable, functional, “good enough” code that I classify as Tier-F.