I believe there are entire companies right now under AI psychosis

Original Article ↗ Hacker News Discussion ↗

Widespread enthusiasm for AI coding tools is colliding with worries about software quality, safety, and loss of engineering discipline. Commenters describe companies mandating heavy AI use, relaxing review and release practices, and justifying shipping buggy code on the assumption that agents can fix issues faster than humans, potentially creating opaque, brittle systems and accumulating “cognitive debt.” Others report real productivity gains when AI is used carefully within strong engineering processes, arguing the technology is already better than many average developers and that the real problem is leadership chasing hype and metrics rather than setting clear boundaries and responsibilities.

What “AI psychosis” is (and isn’t)

  • Many interpret it as companies outsourcing judgment to AI and rationalizing reckless behavior (“agents will fix bugs later”).
  • Others say that’s just another hype‑cycle / cargo‑cult phase, not literal psychosis, and object to misusing a clinical term.
  • Some distinguish between mass groupthink / reality distortion vs genuine AI‑induced psychotic episodes (chatbot delusions, parasocial relationships).

MTTR vs MTBF mindset and software quality

  • Analogy: shift from optimizing “don’t fail” (MTBF) to “recover fast” (MTTR) in cloud ops.
  • Concern: leaders now apply this to AI‑written code — ship quickly, let agents patch production — ignoring hard‑to-detect, long‑running or data‑corrupting bugs.
  • Several argue bug metrics must be “defects introduced per defect fixed”; speed alone is meaningless or harmful.

Experiences with AI coding tools

  • Positive side:
    • Many report 2–5x speedups for small tools, reports, scripts, refactors, and test writing.
    • LLMs can explain complex code, surface subtle bugs, and help cross large codebases.
    • Some teams claim stable or improved incident rates with AI‑assisted workflows plus strict code review and tests.
  • Negative side:
    • “Vibe coding” produces verbose, incoherent, redundant code with hidden coupling and architectural drift.
    • LLMs often “look busy”: plausible fixes that don’t change behavior, ignore specs/tests, or reintroduce bugs.
    • Test suites and “100% coverage” generated by AI can be shallow and miss real defects.
    • Several anecdotes of AI‑rewritten libraries and APIs becoming less reliable.

Management, incentives, and forced adoption

  • Reports of executives mandating “AI everywhere,” measuring token usage, requiring AI in every repo, and pushing “AI-only code review.”
  • Some employees fake AI usage or generate meaningless work to hit AI KPIs.
  • AI is used to justify layoffs and pressure remaining staff to do more with less.

Long‑term risks: debt, security, cleanup

  • Fear of massive cognitive/technical debt: codebases so complex no human understands them, with defect‑fixing agents eventually net-negative.
  • Security worries: AI‑written slop, AI‑poisoned dependencies, and prompt‑injection risks in agentic systems.
  • Expectation that “AI rescue consulting” / “AI janitors” will emerge to clean up failed AI‑built systems.

Profession, economics, and culture

  • Split between “AI already better than average devs” and “AI magnifies mediocrity 10x faster.”
  • Concern that junior devs won’t learn fundamentals if they start by prompting rather than coding.
  • Some hope this crisis will push software toward real engineering discipline; others think short‑term greed will win.