I’m spending months coding the old way

Original Article ↗ Hacker News Discussion ↗

Role of AI vs “coding the old way”

  • Many see value in deliberately coding without LLMs for learning, satisfaction, and maintaining a tight mental model of the codebase.
  • Others argue the highest leverage now is mastering AI/agent workflows; claim large productivity gaps between developers who do and don’t.
  • Some doubt “AI skills” will stay valuable as tools become more user-friendly; others say even basic agent orchestration is hard and will matter for years.

Skills, juniors, and fundamentals

  • Strong concern that new grads who never build complex systems by hand are getting hired as seniors and pushing LLM code straight to production.
  • Several argue juniors must first learn by reading/writing/debugging code manually, otherwise they’ll be helpless when AI fails.
  • Counterpoint: historically juniors were always less effective; industry may simply need far fewer of them as AI productivity rises.

Debugging, perseverance, and cognition

  • Older devs recall multi-day or multi-week debugging as formative; fear LLMs encourage giving up after minutes, weakening persistence and problem‑solving skills.
  • Others see long debugging sessions as wasteful romanticism when an LLM can often narrow down issues in minutes.
  • Multiple comments stress that if devs never build their own debugging muscles, they won’t cope when LLMs can’t diagnose convoluted real‑world bugs.

Agentic workflows, autocomplete, and code ownership

  • Work patterns vary: some use full “agentic” workflows (spec-driven, agents editing repos), others prefer autocomplete-only or “AI as reviewer/rubber duck.”
  • Worries that agent-written codebases become “vibe-coded” systems no one truly understands, increasing technical debt and risk.
  • Some refuse to accept AI-generated core code at all, only using LLMs for explanations, research, or review to preserve ownership and understanding.
  • Several praise autocomplete as the best middle ground; others find it underwhelming compared to agents.

Education, low-level work, and retro computing

  • Examples of teaching 6502 assembly with line editors: pain initially, but students start planning more and holding programs in their heads.
  • Some see similar benefits in coding by hand today: better abstraction skills, deeper grasp of types and architecture.
  • There’s speculation about AI-resistant teaching languages or environments, but feasibility is unclear.

Career and industry outlook

  • Some predict SWE roles, especially junior positions, will shrink dramatically; others note ongoing demand in security, infra, and safety‑critical domains.
  • Broad unease that widespread “vibe-coded” systems plus weakened human skills could lead to severe technical debt and future crises.