Where is all the new, good software?

Original Article ↗ Hacker News Discussion ↗

Rapid advances in AI-assisted coding have not yet produced a visible wave of “new, good software” for most users, prompting debate over where productivity gains are actually going. Many engineers report that code generation is faster but decision-making, review, QA, organizational bottlenecks, and risk aversion limit real-world impact, while overall software quality may even be declining. Others argue that AI is enabling a quiet boom in highly personalized, one-off tools and niche projects that rarely reach app stores or broader markets, making the benefits harder to see from the outside.

Scope of AI Productivity Gains

  • Many argue AI accelerates coding but not overall delivery. Bottlenecks are design, decision-making, coordination, ops, QA, and review.
  • Several breakdowns suggest even perfect automation of coding/debugging yields maybe ~1.25–2x productivity, not 10x.
  • Code review and understanding often get slower because AI produces more, and sometimes jankier, code to scrutinize.
  • Claims that teams are relaxing and taking “coffee breaks” are widely rejected; commenters report layoffs, higher pressure, and no extra leisure.

Software Quality: Better or Worse?

  • Numerous anecdotes of more bugs, worse performance, and fragile AI-generated code (especially “vibecoded” apps and AI-first products).
  • Others report clear internal improvements: more tests, fewer Sentry issues, better performance, faster bug-finding using agents and LLMs.
  • Consensus: disciplined teams can get quality gains, but many orgs lack the rigor or QA staffing to realize them.

Where Is the “New, Good” Software?

  • Visible “new” software is often AI-built clones (Photoshop, Excel, PDF readers) perceived as alpha-quality, slow, and incomplete.
  • Some see these clones as promising early alternatives to monopolies; others call them “slop” and morally dubious copycats.
  • Several argue the “project of software is mostly complete”: most broad categories exist; new work is integrations, niche tools, or incremental UX.

Bespoke and Invisible Software

  • Many participants have built high-quality personal tools (accounting apps, home automation, 3D house planners, readers, synth firmware, dev utilities).
  • These are often better for their creators than commercial options but are not released due to:
    • Marketing/distribution being a grind.
    • No clear rewards, plus long-term maintenance burden.
    • Hostile reception and “AI slop” stigma.
  • Result: lots of “new, good” software exists privately on individuals’ machines.

Ecosystem, Incentives, and Discovery

  • App stores feel closed; the “old internet” as a discovery channel is described as dead or drowned in low-quality AI projects.
  • Most trending repos are AI tooling, agents, and orchestration, not end-user apps.
  • Organizational incentives favor more features and scope over higher quality; teams are kept at “bare minimum” staffing.
  • Some foresee a “fast-food era” of software: more ephemeral, defect-tolerant tools shipping faster, with quality de-emphasized.