Vibe engineering

Original Article ↗ Hacker News Discussion ↗

What “vibe engineering” is trying to capture

  • Proposed as a term for structured, test‑driven, AI‑assisted development, to distinguish it from “YOLO” vibe coding where code isn’t read or really understood.
  • Described as closer to managing a team of over‑eager junior devs: planning, specs, tests, CI, architecture, and constant review, not just prompting and pasting.
  • Some argue this is simply software engineering with new tools, not a new category.

Perceived benefits of LLM/agentic workflows

  • Many experienced devs report large gains for:
    • Prototyping multiple approaches quickly, especially in unfamiliar stacks.
    • Refactors, boilerplate, test scaffolding, and brownfield cleanup.
    • “Spec‑driven” loops where the model helps write and refine specs, then implements them.
  • Tools like AGENTS.md/CLAUDE.md, small custom scripts, and strict TDD are seen as key to getting reliable output.
  • Good existing engineering practices (version control, tests, CI, documentation, architecture) act as “guardrails” that make agents far more useful.

Skepticism, risks, and limitations

  • Several career devs report <10% benefit or even slowdowns: subtle bugs, review fatigue, and time lost steering confused agents.
  • Concern that people overestimate productivity gains based on anecdotes; calls for better empirical studies.
  • Fear of “AI slop”: verbose, inconsistent, hard‑to‑maintain code, especially in large legacy codebases.
  • Worries about cargo‑cult “superstitions” around prompting, changing models frequently, and lack of reproducibility.

Process, quality, and liability

  • Broad agreement: without strong tests, linting, and clear specs, agent code is dangerous.
  • Some insist “engineering” implies personal accountability and liability, which current AI‑heavy workflows don’t meet—especially outside software (bridges, hardware, safety‑critical systems).

Impact on developers and the profession

  • Many describe discouragement: the craft feels like it’s turning into managing unreliable agents instead of hands‑on coding.
  • Others say LLMs amplify senior engineers’ leverage, widen the gap with juniors, and enable solo devs to ship much more.
  • Debate over long‑term effects: skill atrophy vs. freeing humans to focus on higher‑level design and problem selection.

Debate over naming

  • Strong dislike of “vibe” as unserious or pejorative; alternative labels suggested include “agentic coding,” “AI‑assisted programming,” “augmented engineering,” and “agent‑assisted coding.”
  • Some think no special term is needed: it’s just software engineering with better tools.