The Cotton Gin Question
Will AI create more software work than it destroys? A first-principles investigation of the Jevons Paradox in software development.
The 19th century saw two very different technologies hit the cotton industry. The cotton gin made processing cheaper, demand exploded, and the industry (and its labor force) grew for decades. A century later, the mechanical cotton picker made harvesting cheaper too, and that labor category simply vanished within years. Same industry, same logic of "make it cheaper," opposite outcome for workers.
AI is doing to software what both machines did to cotton, at the same time. This paper works through the historical analogies, the economics, and seven competing theories to answer the question every developer, founder, and policymaker is actually asking: is coding heading for the gin's outcome, or the picker's?
The core finding
- 01
Total software produced will almost certainly explode (~90% confidence) — falling prices unlock vast latent demand from the world's ~350M underserved small and mid-sized businesses.
- 02
Whether human developers benefit is a separate, harder question — coding was never the majority of software's lifecycle cost. The bottleneck is migrating to specification, verification, and trust, not disappearing.
- 03
The likely labor market is a barbell: fewer traditional mid-level coding jobs, more highly paid verifier-architects at the top, and an explosion of non-professional "citizen builders" at the bottom — the same pattern spreadsheets created in the 1980s.
That 25-point gap between the two numbers is the paper's central thesis: more software-related work overall, distributed very differently across professions than it is today.
Seven competing theories, rigorously scored
Each theory is tested against historical precedent, economic reasoning, and its strongest counter-case, then assigned a probability:
Plus a wildcard scenario — a monoculture cascade, where AI-generated code creates systemic vulnerabilities and a correlated mass failure paradoxically triggers a boom in senior-engineer hiring (15% likelihood).
Why it matters beyond software
The paper's framework generalizes past coding: any knowledge-work profession facing AI automation should ask the same question. Is the bottleneck the mechanical task itself, or is it trust, verification, and judgment sitting downstream of that task? That's the question that determines whether a given profession follows the cotton gin or the cotton picker.
Read the full paper — historical case studies, the economic framework, all seven theories in full, ten new software categories AI enables, and a 12-point indicator dashboard to track which theories are winning.
Download the full paper (PDF) ↓This paper was researched and drafted collaboratively by a human and an AI model, with the collaboration itself credited transparently in the byline. If you want to talk about what this means for your business, or you think one of these seven theories is wrong, get in touch.