Industrialized Software and the Coming Stewardship Crisis

AI is driving an industrial revolution in software, creating cheap, scalable ‘disposable software’ and accelerating output via Jevons paradox. While some human-crafted work will persist, lasting value will come from innovation, with industrialization rapidly commoditizing new capabilities. The looming crisis is stewardship: technical debt and maintenance for software that no one truly owns.

Key Points

• AI coding is industrializing software production, enabling cheap, scalable, and less expert-dependent output—including a new class of disposable software.
• Jevons paradox suggests efficiency will increase total software consumption, likely flooding the ecosystem with low-value, high-volume artifacts (“AI slop”).
• Analogies to other industries (ultraprocessed foods, user-generated media) show how industrial economics drive quantity over quality—and why the trend is sticky.
• Innovation and industrialization are distinct but complementary: innovation expands the solution space; industrialization commoditizes and scales it. LLMs mark an inflection point accelerating this cycle.
• Industrial dominance brings externalities—dependency sprawl, maintenance burden, and security risk—making stewardship and maintenance the central, unresolved problem.

Sentiment

Overall, the sentiment of the discussion is largely skeptical and critical of the article's core premise and its analogies. While there's a recognition of AI's impact on increasing automation and speeding up coding, many commenters disagree with the notion that software is not already industrialized or that industrialization inherently leads to lower quality. The unique economic properties of software (zero marginal cost) are frequently cited to challenge the article's comparisons to physical goods, suggesting a more nuanced outcome than a simple shift to 'disposable slop.'

In Agreement

• AI will accelerate the automation of code generation and development processes, speeding up parts of the software creation lifecycle.
• The concept of 'disposable software' or highly customized, temporary niche applications (often for personal or small-scale use) will become more feasible due to AI, solving problems previously uneconomic.
• The analogy of 'technical debt as pollution' is highly relevant and effectively highlights the growing systemic risks like sprawling dependency chains, maintenance burdens, and expanding security surfaces.
• Quality can often be sacrificed for speed to market, making 'low-quality slop' a viable outcome for some applications in a competitive landscape.
• LLMs are a significant accelerant in the ongoing trend of software industrialization, even if not an entirely new revolution.

Opposed

• Software development is *already* highly industrialized, with widely used applications (browsers, OS, databases) produced by large, specialized teams using advanced tooling and CI/CD, making the article's premise outdated.
• The analogy between physical goods (clothing, agriculture) and software is flawed because software has zero marginal cost, leading to different market dynamics where quality often trumps price, and industrialization can improve rather than degrade quality.
• LLMs primarily speed up code *writing* (the easy part) but do not effectively address the complex, 'slow parts' of software development like design, architecture, requirements gathering, quality assurance, and compliance, which require taste, insight, and human expertise.
• Industrialization does not inherently lead to low-quality products; it can also enable precision and consistent quality far beyond artisanal methods (e.g., CNC manufacturing, modern cars).
• Software's value often comes from decades of accumulated feedback, tuning, fixing, and battle-testing, which AI cannot easily replicate or accelerate, making competition with established products difficult.
• The article's framing of industrialization as inherently leading to 'junk' and its analogies are sometimes seen as disingenuous or sensationalist, oversimplifying economic and historical realities.
• LLMs' output often requires significant human 'captaining,' debugging, and correction, especially for complex or specific requirements, making the process faster but often frustrating rather than truly autonomous.
• The 'Jevons Paradox' might have limited scope in the context of LLMs, as some research suggests the price elasticity for their use is not high enough to cause massive, unbounded consumption of generated software.