Is AI ruining our skills? Early results are in – and they're not good

Original Article ↗ Hacker News Discussion ↗

Early research and developer anecdotes suggest that heavy reliance on AI assistants may erode core skills, from medical image reading to everyday software engineering and even basic reasoning and communication. Commenters debate whether this is simply the latest wave of tool-driven “skill atrophy” (like calculators or GPS) or something qualitatively different because LLMs touch nearly all cognitive work, potentially dulling the expertise needed for innovation and judgment. Many see a trade-off emerging: broader access and higher productivity for novices and managers, at the possible cost of deep technical mastery, critical thinking, and long-term career resilience.

Perceived Skill Atrophy

  • Many participants report firsthand decline in skills (coding, planning, problem‑solving, even typing) after sustained AI use, often describing a “relearning to walk” phase when they stop.
  • Core worry: not just loss of narrow skills (like arithmetic or CLI commands) but erosion of broad reasoning, judgment, and “blank‑page” problem solving.
  • Several note that AI use can undermine users’ ability to evaluate AI output, creating a feedback loop of dependence.

Tool Use, Trade-offs, and Historical Analogies

  • Some compare LLMs to calculators, compilers, cars, or Ghost disk imaging: tools that make old skills unnecessary and let new layers of abstraction emerge.
  • Others argue the difference is scope: LLMs touch “basically all knowledge and communication skills,” not a narrow domain.
  • There’s general agreement that unused skills atrophy; dispute is whether that’s acceptable or existentially dangerous.

Impact on Software Engineering Practice

  • Multiple anecdotes of senior engineers “vibe‑coding,” shipping more but with worse code and weaker judgment.
  • Concerns that high-end, nuts‑and‑bolts expertise (systems, languages, compilers, architecture) will shrink, risking a slowdown in genuine innovation even as CRUD output explodes.
  • Reviewers struggle to mentor juniors because AI-written code no longer reflects the junior’s own thinking.
  • Others report the opposite: system design and architectural thinking improving because AI handles boilerplate and enables faster exploration and refactoring.

Learning, Education, and Cognitive Effects

  • Split views:
    • Some use AI as a tutor to tackle hard subjects (physics, quantum mechanics), new languages, and physical skills, emphasizing drills and verification.
    • Others say this usually leads to superficial “edutainment,” wide but shallow understanding, and kids doing homework by pasting prompts with zero reflection.
  • Breadth vs depth is a recurring theme: easy curiosity scratching vs hard, generative mastery.

Open-Source, Access, and Power Concentration

  • Debate over whether open models are “close enough” to frontier systems and whether that’s sufficient once a model is “good enough” for a task.
  • Worries that if only a few large providers control top models, society becomes cognitively dependent on a centralized, political, or corporate “oracle.”

Workplace Incentives and Career Dynamics

  • Many note that corporate incentives (velocity, cost-cutting) favor aggressive AI use, pushing deskilling while demanding more output.
  • Some foresee management paths opening sooner for mediocre but AI‑amplified developers; others predict a premium on those who retain deep skills.
  • There’s tension between short‑term productivity gains and long‑term professional and societal resilience.