Web development is fun again

Original Article ↗ Hacker News Discussion ↗

Role of LLMs in Making Dev “Fun Again”

  • Many describe LLMs as removing the “activation energy” for side projects: you can get a working prototype or plugin in under an hour, where previously you’d need multi-hour ramps.
  • Parents, managers, and ex-developers say they’re coding again because they can make progress in small, sporadic time slots.
  • Common pattern: describe the desired system, let an agent scaffold code, then review/tweak. AI is often used for boilerplate, wiring APIs, build configs, CSS/flexbox, tests, docs, and dependency upgrades.

Process vs Outcome: What’s Actually Fun?

  • Strong split:
    • One camp enjoys results (a working tool, bot, UI) and is happy to offload typing and “plumbing” to AI.
    • Another camp enjoys the craft of programming; for them, prompting feels like hiring a trainee to do the hike or cook the meal—less satisfying.
  • Long analogies (cooking/head chef, hiking/boots, IKEA vs carpentry, printing press) are used to argue over whether AI-assisted work “counts” as programming.

Quality, Slop, and Maintainability

  • Supporters: AI can be like a powerful junior dev or power tool—great if you already know what “good” looks like and review/test accordingly.
  • Skeptics: “vibe-coded” codebases accumulate technical debt and weird failure modes; inheriting such systems without AI is frightening.
  • Many worry about sloppier engineering hygiene, over-trust in agents, and markets flooded with low-effort clones.

Specialization, Roles, and Skills

  • One view: LLMs “bail us out” of extreme specialization and empower generalists to span more domains (web, embedded, infra, etc.).
  • Counterview: they commoditize generalist output; only deep specialists retain clear market value.
  • Debate whether PMs/systems analysts using AI are “back in development” or doing something fundamentally different from traditional software engineering.

Web Stack Complexity vs Simpler Stacks

  • Many see LLMs as coping mechanisms for a needlessly complex frontend ecosystem (bundlers, frameworks, build chains).
  • Others insist you can still have fun with Rails, PHP + SQL, SSR, HTMX/HARC, Alpine, Tailwind, or plain HTML/CSS/JS—LLMs optional.
  • Some argue the complexity was always there; we just see and manage more of it now.

Learning, Productivity Claims, and Dependence

  • Mixed reports on actual multipliers: personal anecdotes range from modest speedups to “10x” for side projects, but some find agents slow, brittle, or cognitively dulling.
  • Concerns include: reduced deep learning, over-reliance on opaque tools, subscription lock-in, and future “AI withdrawal” if services disappear.
  • Others counter that they’ve learned more—especially in unfamiliar stacks or languages—by studying AI-generated code and iterating on real projects.