Vibecoding isn't as fun as writing code by hand
Many long-time programmers say letting AI agents write most of their code feels less satisfying than “hand coding,” trading the joy of flow, craftsmanship, and learning for a more managerial role of reviewing and steering machine output. Others welcome so‑called “vibecoding” as a power tool that removes boilerplate, unlocks previously infeasible side projects, and lets domain experts build software without deep programming skills. Across perspectives, people wrestle with how much they value the process versus the end product, how to keep skills sharp in an AI‑heavy workflow, and what this shift means for careers in software development.
Subjective “fun” and how to frame it
- Many argue fun is inherently personal; saying “X isn’t fun” is obviously about one’s own experience.
- Others think authors should still signal subjectivity (“I think”, “for me”) to avoid endless argument and clickbait perceptions.
- Some see the backlash as defensiveness from AI users; others see it as annoyance at being told what they “should” enjoy.
How people actually use AI to code
- Wide spectrum: from “no-code reading, just chat and ship” to “AI as autocomplete / refactoring assistant” to full “agentic” setups with planning, coding, and review agents.
- Common pattern: hand‑write interesting or critical parts, delegate boilerplate, glue code, or one‑off integrations.
- Some use AI mainly for small personal utilities or resurrecting old projects; others for large professional systems, with metered usage in enterprises.
Perceived benefits of vibecoding
- Huge speedups, especially for: boilerplate, complex-but-standard stacks, APIs you’ll only use once, tedious debugging, converting or modernizing old code.
- Lets “idea people” or domain experts who aren’t strong coders build things they previously had to hire out.
- Many long‑time devs report feeling like “architects” or “conductors” able to realize long‑held ideas quickly.
Perceived downsides and loss of joy
- Strong sense of lost “flow” and craftsmanship; work feels more like management, DevOps, or code review than creation.
- Vibecoding can feel like pulling someone else’s code from the internet, or “watching someone play the game instead of playing it.”
- Some find the dopamine of solving hard problems replaced by slower, less satisfying organizational and planning problems.
- Several plan to leave or de‑emphasize professional coding because the fun part is gone.
Code quality, risk, and review burden
- Skeptics: LLMs amplify average/mediocre code, can ignore existing abstractions, and introduce subtle security or reliability risks; risk‑sensitive customers care.
- Supporters: with good architecture, prompting, and multi‑agent review, output can match or beat typical human code, and sheer volume plus fast iteration is a major advantage.
- Many complain about the “review dilemma”: review everything (burnout) vs. trust the agent (production surprises).
Careers, identity, and gatekeeping
- Two self‑described “camps”: those who love code as craft vs. those who primarily care about end products and users.
- Tension: some accuse critics of “cope” or gatekeeping; others accuse heavy AI users of laziness or accepting “slop.”
- Widespread anxiety about future employability, especially for people who mainly enjoy coding itself.
Learning to code in the AI era
- Older devs advise beginners: use AI as tutor/reviewer, but type your own implementations if you want deep understanding; “there is no passive learning.”
- Concern that over‑reliance will erode skills and make large AI‑written codebases hard even for AI to maintain.
- Others emphasize a new skill gap: knowing how to drive agents effectively and architect around them.