AI code and software craft

Enterprise vs Consumer Software Incentives

• Enterprise tools are often bad not just because buyers don’t use them, but because big-paying customers demand bespoke features and weird configuration paths that outlive their original sponsors.
• Consumer software can be more polished but is often optimized for engagement, not actual value.
• Misaligned incentives (manager vs frontline worker) create friction: managers want data and controls; workers see slow, annoying UIs with duplicate data entry and no budget for proper integrations.

AI as Industrialization: Luddites, Cloth, and Quality

• Several comments recast the debate as a modern Luddite vs industrialist conflict: craft/agency vs efficiency/scale.
• Others push back: early industrial cloth and many modern garments are argued to be worse (and more environmentally harmful) even if cheaper and more abundant; quality decline is framed as both an engineering constraint and an economic choice.
• Parallel drawn: even if AI output is worse, it can still displace human labor, just as lower-quality machine-made goods did.

Craft, Plumbing, and What Most Software Really Is

• Many argue most industry software is already “plumbing” and largely mediocre; AI simply matches that baseline and exposes how little “craft” was happening anyway.
• For some, AI tools finally make it feasible to ship side projects and experiments that previously died at the “init commit” stage.
• Others counter that the idea AI will “free up” engineers to do more craft is wrong; instead it may finish off what remains of craftsmanship, relegating hand-coding to a niche hobby, like blacksmithing.

Code Quality, Correctness, and AI Slop

• Strong divide on AI code quality: some say agents can produce high-quality code with orchestration, tests, and review; others say generated code is “orders of magnitude worse” and creates huge, hard-to-verify diffs.
• Consensus that AI is great for boilerplate, glue, scaffolding, and small internal tools; much weaker for system-level reasoning (auth boundaries, failure modes, state consistency).
• Several note AI amplifies existing tendencies: good engineers get faster; sloppy ones produce more slop.

Labor, Training, and Incentives

• Concern that if one senior can do the work of multiple juniors with AI, companies will stop hiring juniors, hollowing out the pipeline of future experts.
• Others liken this to offshoring and open source: long-running forces that already devalued some aspects of coding labor.
• A few insist the real problem is incentive structures: productivity gains are being used to cut headcount, not buy humans time or improve quality.

Control, Understanding, and Tooling Limits

• Debate over how much “control” developers truly have over LLMs: some claim you can strongly steer architecture and style; critics say you only influence probabilities and must constantly guard against models “going off the rails.”
• Disagreement over whether current systems “understand” anything; some see that critique as philosophical hair-splitting if the tool is practically useful for software tasks.

Societal and Political Concerns

• One branch worries AI-generated media will so thoroughly pollute the information environment that people no longer trust any event, neutering mass mobilization and accountability.
• Others argue media credibility was eroding already; AI is another accelerant but might also force long-overdue investment in identity, trust, and security.

Efficiency, Metrics, and the Fate of Craft

• Multiple comments connect AI’s rise to a broader cultural fixation on efficiency as the supreme value, even when it undermines resilience or long-term health.
• Because efficiency and output are easy to measure and “craft” is not, organizations naturally optimize for the former—AI fits neatly into that logic.
• Some remain hopeful that while AI will flood the world with “slopware,” the absolute amount of well-crafted software might still grow, created by those who deliberately use these tools to extend, not replace, human judgment.