AI makes programming differently difficult

Original Article ↗ Hacker News Discussion ↗

AI coding tools are widely seen as automating the “easy” parts of programming—syntax, boilerplate, and common patterns—while shifting the hardest work toward judgment, architecture, and understanding trade-offs. Commenters argue this both raises the bar for effective use (experience is needed to validate and steer AI output) and accelerates productivity for seasoned developers, fueling anxiety about job displacement, code quality, and how newcomers will learn foundational skills. Many expect programming roles to change substantially rather than vanish, with long-term value concentrated in problem framing, system design, and integrating machine-generated code into maintainable systems.

Overall effect of AI on programming difficulty

  • Many argue AI makes programming easier by automating syntax, boilerplate, glue code, refactors, and rote YAML/config work.
  • Others say difficulty has shifted, not decreased: from recalling APIs and writing code to judging plans, designs, and correctness.
  • Some feel architecture and “thinking clearly at scale” were always the real hard part; AI just exposes that by removing the easy parts.
  • A few claim AI already writes better code than they do for many tasks, turning their role into planning, QA, and architecture rather than typing.

Architecture, judgment, and mental models

  • Recurrent theme: AI can’t yet handle complex architecture, long-horizon coherence, or nuanced trade-offs without strong human direction.
  • Good results require up-front, detailed design and strong “rails”; this can feel harder than evolving designs while hand-coding.
  • The hard part now is evaluating whether AI-generated plans and code actually make sense, which presupposes substantial prior experience.

New skills and workflows

  • New skills cited: writing precise specs, feeding context effectively, orchestrating multiple agents, recognizing when AI is failing, and knowing which tasks to delegate.
  • Several note decision fatigue: reading long AI plans and making many small calls is mentally exhausting.
  • Effective workflows often involve: use AI for first drafts (code or docs), then heavily edit or even rewrite once intent is clarified.

Code quality and failure modes

  • AI code is praised for consistency, documentation, tests, and applying known patterns; excellent for “boring” infra and CRUD-style tasks.
  • Criticisms: confident wrongness, hidden bugs, over-engineering, messy global structure, and repeated reinvention of existing functions.
  • AI-generated text and code can look polished yet lack real insight or solid underlying structure, making it deceptively hard to review.

Learning, seniors vs juniors, and tacit knowledge

  • Many believe experienced developers gain the most, because they can spot nonsense and guide agents; juniors may struggle to learn judgment when AI hides complexity.
  • Tacit knowledge, systems thinking, and knowing “what not to build” are seen as hard to encode in training data and remain human strengths.

Jobs and long-term outlook

  • Views range from “end of programming as a mass, well-paid profession” to “another abstraction layer that still needs humans on top.”
  • Some expect rapid progress leading to near-total automation; others predict plateaus and continued demand for humans to direct, diagnose, and clean up AI-built systems.