The rise of industrial software

Has Software Already Been “Industrialized”?

• Some argue software’s “industrial revolution” happened long ago with high‑level languages, reusable components, containers, and cloud.
• Others say current LLM tools (Claude, Codex, etc.) are only the beginning of a much steeper curve in productivity and scale.
• A third view is that most “industrialization” happened in the 60s–70s; LLMs mainly accelerate an already‑industrial process rather than inaugurate a new one.

Industrialization Analogies: Where They Fit / Break

• Critics say the article cherry‑picks downsides of industrialization (junk food, fast fashion) while omitting huge gains in availability, quality, and longevity of many mass‑produced goods.
• Several point out that software differs from physical goods: zero (or near‑zero) marginal cost, instant copying, and no inherent link to population or wear.
• Others think the “industrialisation” framing is still useful as a metaphor for plummeting production costs and explosion of low‑value output, even if the economics differ.

Quality, Junk, and “Disposable” Software

• Many doubt there’s broad demand for disposable apps; businesses want secure, durable, maintainable systems.
• Some see a niche: tiny, one‑off tools (“glue” between fragmented systems, personal automations, kids’ joke apps) where throwaway code is exactly right.
• Others note software was already mostly non‑artisanal; the tsunami of mediocre software just gets larger and cheaper.

Economics, Demand, and Marginal Cost

• Debate over whether economic growth tracks energy use and whether AI‑driven growth hits physical limits.
• Several stress that in software, prices are already free or “dirt cheap,” so cheaper development doesn’t create a new low‑cost market segment the way industrial goods did.
• Some expect AI mainly to unlock small, previously uneconomic niches (custom tools for small businesses, nonprofits, families).

LLMs in Practice: Capability and Limits

• Reports of “vibe‑coded” projects: LLMs speed scaffolding and glue code but still need a “captain” with domain understanding and design taste.
• Experienced devs say LLMs help with speed, searches, refactors, and porting algorithms, but don’t yet manage complexity, architecture, or requirements.
• Skeptics say productivity gains are overstated; for anything nontrivial, it’s still faster and safer to code manually.

Knowledge, Interfaces, and Lock‑In

• Several highlight user learning cost as missing from the essay: changing UIs drains a “knowledge pool” and forces retraining.
• Some tie this to open‑source cultures that prioritize stable interfaces (e.g., Unix tools, traditional editors) vs ecosystems with frequent breaking changes.

Maintenance, Technical Debt, and Stewardship

• The “technical debt as pollution” metaphor resonated with some: mass automation amplifies hidden maintenance and security costs.
• Others counter that good teams consciously manage debt; it grows when organizations rush and misunderstand domains.
• Broad agreement that stewardship—who maintains vast quantities of semi‑ownerless code—remains unresolved, especially if LLMs flood ecosystems with more fragile software.