Learning to code is still worthwhile
Whether it’s still worth learning to code in the age of powerful AI models is increasingly contested. Many argue that programming remains essential for understanding and controlling complex systems, exercising critical thinking, and creating reliable, maintainable software—even as LLMs automate more “outer layer” coding and amplify both good and bad developers. Others counter that AI will shrink demand for average programmers, turning coding into a more niche or artistic pursuit, and urge newcomers to weigh the long-term career risks against the intrinsic benefits of the skill.
LLMs and the Future of Coding
- Many argue current and near‑future LLMs greatly reduce the need to hand‑write routine code, especially for CRUD apps and “outer layer” business software.
- Others counter that LLMs are non‑deterministic, brittle, and poor at high‑level abstraction, so skilled programmers remain crucial, especially for core infrastructure, compilers, frameworks, and safety‑critical systems.
- Some foresee a small elite of deep experts overseeing powerful AI tools, with far fewer traditional coding jobs; others expect Jevons‑style effects where cheaper software creation increases total demand.
Why Learning to Code Still Matters (According to Supporters)
- Teaches problem decomposition, debugging, and systematic thinking; often compared to learning math or philosophy for reasoning skills.
- Essential to judge, guide, and constrain LLM output; “prompt‑only” users can’t reliably tell good solutions from bad.
- Foundational understanding (from logic gates up to high‑level languages) helps when abstractions leak and AI can’t fix edge‑case failures.
- Seen as a way to fully engage with and shape AI systems rather than passively delegating everything.
Skepticism: Is It Still a Good Career Bet?
- Some compare future programming work to poetry, early music, or niche crafts: rewarding but economically precarious.
- Concern that LLM‑driven productivity improvements will reduce demand for average coders, amplifying only top performers.
- Worries about advising new students to take on debt for a field that may be heavily commoditized within 5–20 years; analogies to blacksmiths, tool‑and‑die workers, and outsourced IT.
Abstraction, Fundamentals, and Understanding
- Broad agreement that “learning to code” is really about understanding systems, not memorizing syntax.
- Debate over how deep one must go (assembly, CPU design, etc.), but many fear AI will encourage shallow, cargo‑cult knowledge.
Coding as Craft, Art, or Plumbing
- Some see code as a creative medium on par with music, literature, or painting; others say most industry work is closer to plumbing or carpentry—necessary, constrained craft.
- Several note that LLMs risk stripping away the “flow” and joy of hand‑crafting abstractions, turning senior work into reviewing and “babysitting” machine‑generated code.
Quality, “Slop,” and Long‑Term Risks
- Reports that weak developers still produce bad code with LLMs, just faster; good developers can learn faster and explore designs more.
- Fears of an ecosystem filled with opaque, bloated, AI‑generated code that humans (and later models) struggle to maintain.
- Minority view claims AI‑generated code is already often cleaner than typical human code, challenging assumptions about quality decline.