Vibe coding has turned senior devs into 'AI babysitters'

Original Article ↗ Hacker News Discussion ↗

Project guidance & AGENTS files

  • Several commenters recommend an AGENTS.md (or similar) file with security rules, coding style, test expectations, and project context to steer AI agents.
  • Some link to repos of template AGENTS files and say tools can auto-generate an initial version (/init).
  • Others doubt this helps in typical teams where many devs already resist writing or reading documentation, suggesting even AGENTS docs might need their own generator.

Documentation quality debate

  • One camp: AI’s need for context could finally incentivize better docs, structured repos, and guardrails that also benefit humans.
  • Counterpoint: we’ll get more docs, not necessarily better ones; business context and use-case documentation remain hard.
  • Another view: AI-generated docs are significantly better than old comment-extraction tools because they can reason over callers/callees and implementation.

Vibe coding, PR slop, and team norms

  • Strong frustration with “vibe-coded” PRs: large AI-generated changes that are obviously wrong, poorly tested, or labeled [vibe] as a soft disclaimer.
  • Multiple people advocate normalizing hard rejection of low-effort AI PRs and holding authors fully responsible regardless of AI use.
  • Some suggest first-pass AI code review, but others say AI reviewers always “find something,” making them noisy and unhelpful.
  • Several anecdotes describe huge, unnecessary AI-driven changes that waste senior time to unwind.

Productivity, prompting skill, and AI babysitting

  • Experiences diverge sharply. Some say AI lets them operate more like architects, quickly executing multi-file refactors or boilerplate.
  • Others feel they spend more time crafting prompts and supervising than just coding, and that reviewing AI output is boring and demoralizing.
  • One perspective frames this as a leadership/management problem: working with AI is like managing many novice juniors; success depends on planning, prompts, and knowing when to interrupt.
  • Another thread argues “prompt engineering” can yield weeks of work from a single well-crafted prompt; skeptics press for concrete examples.

Workforce, incentives, and media

  • Observations that companies are cutting juniors and leaning on AI plus seniors, effectively “offshoring” work to LLMs and risking long-term expertise.
  • Concerns that “AI babysitting” turns seniors into cleanup crews for hype-driven decisions, with comparisons to gold-rush “shovel sellers” and prior outsourcing/H‑1B cycles.
  • Several express fatigue with AI boosterism in media and from tool vendors, feeling the discourse is dominated by marketing.