Now might be the best time to learn software development

Overall reaction to the essay

• Many found the piece funny, refreshing, and a useful antidote to AI doomerism about developers.
• Some disagreed with specific framings (e.g., scale of productivity improvement) but broadly agreed that AI shifts the role of developers rather than erasing it.

Historical analogies: Fortran, COBOL, SQL, “no‑code”

• Commenters drew parallels to past “anyone can program now” waves: Fortran, COBOL, SQL, QUEL, UML tools, FrontPage, Delphi, Flash, Dreamweaver, Excel, node-based tools.
• Pattern noted: tools do raise productivity and broaden who can “program,” but:
  • They don’t eliminate demand for serious developers.
  • They eventually hit inherent limits; hand-written code (or more general tools) outlasts many of them.
• SQL is highlighted as a partial success: many non‑programmers use it, but many developers still struggle and hide behind ORMs.

Farming, photography, and Jevons-style economics

• The combine-harvester analogy split opinion:
  • Some see LLMs as similar: one dev can do the work of many; specialization around the core activity will expand.
  • Others say Fortran was a bigger productivity leap than current AI.
• Debate over Jevons: food demand is partly inelastic, but waste and diet shifts suggest some elasticity.
• Important distinction: farmers historically owned the capital; most developers are employees, so owners may capture more gains.
• Photography analogy:
  • Digital cameras massively increased quantity and lowered entry barriers, creating more good photos and more competition.
  • Professional photographers’ income and job security declined; “good enough” flooded the market.
  • Some see this as a template for software: more apps, more “ok” work, tougher economics for pros.

LLMs in current software practice

• Many report genuine productivity boosts:
  • Navigating large unfamiliar codebases.
  • Generating boilerplate, tests, deployment scripts, and reports.
  • Turning “I know what to do but not where/how” into concrete steps.
• LLM-produced “vibe-coded” apps are already creating cleanup work; Upwork has clients stuck in AI‑generated pits.
• Several compare LLMs to an overconfident junior dev: good at scaffolding, bad at edge cases and refactors; needs supervision.
• Agentic/“fire and forget” coding is widely seen as unreliable; treating AI as an assistant, not an autonomous coder, works better.

Learning, Stack Overflow, and fundamentals

• Many see this as a uniquely good time to learn programming:
  • LLMs fill the old “friendly Stack Overflow” role: tutoring, debugging help, alternate explanations.
  • They make early learning less lonely and more interactive; you can learn by debugging AI output.
• Strong warnings not to skip fundamentals: without mental models (performance, concurrency, security, data structures), you can’t judge or fix AI output.
• Several lament Stack Overflow’s decline and worry about the training data pipeline if fewer people share solutions publicly.

Psychological impact: support vs drain

• Some find LLMs provide valuable “psychological support”: a rubber-duck partner that breaks procrastination and keeps momentum.
• Others find them emotionally draining—“overconfident idiot” or “yes‑man” interactions that require constant correction, with no real pushback or insight.
• People compare this to bad Google results vs. bad AI; both can be exhausting in different ways.

Jobs, wages, and prompt engineering

• There’s visible anxiety: recent layoffs, worse interviews, talk of “productivity and cost reduction” from management.
• Disagreement on actual unemployment levels, but consensus that perception matters more than theoretical productivity arguments.
• Debate over whether prompt engineering will:
  • Become a high‑paid specialization (if value and scarcity are high), or
  • Be low-paid because barriers to entry are low and many can do it.
• Some suggest reskilling outside software as a pragmatic hedge; others argue this is also a rare window for bootstrapped AI-era startups or building portfolio projects.

Labor power, efficiency, and distribution

• Several note that productivity gains don’t automatically flow to workers; they hinge on power, organization, and policy.
• Discussion around unions: tech is largely non-union, making it easier for firms to use AI as a pretext for wage suppression and headcount cuts.
• Others counter that individual resistance (refusing hype, careful tool choice) is possible but weaker than collective action.

Skepticism about long-term AI programming dominance

• Multiple commenters caution that most “this is the future of programming” predictions in history have been wrong.
• Even if LLMs remain useful, particular tools and workflows (like past RAD tools) may fade, so betting an entire career on one AI stack is seen as risky.