The Gorman Paradox: Where Are All the AI-Generated Apps?

Original Article ↗ Hacker News Discussion ↗

Where Are the AI-Generated Apps? (Visibility vs Reality)

  • Many commenters say AI-built apps do exist but are mostly:
    • Internal line-of-business tools (custom CRMs/ERPs, accounting tools, flight school rental systems, infra automation, embedded code).
    • One-off “vibe-coded” utilities, scripts, browser extensions and CLIs tailored to a single user or team.
  • These rarely hit app stores or public marketplaces, so app-store counts miss most of the impact.
  • Others note lots of obvious “AI slop” on the web and in app/game stores (e.g. low-quality Steam games, ugly marketing sites) but agree that high‑quality, widely used AI‑built products are scarce.

What AI Coding Is Actually Good At

  • Strong for: scaffolding CRUD apps, parsing common formats, internal dashboards, scripting, refactors, tests, rote API glue, and debugging when guided by an experienced developer.
  • Several report large personal speedups (often 3–10x) when they already understand the domain and review/shape the code as it’s written.

Where It Still Fails: The Last 20%

  • Major weaknesses:
    • Handling messy real-world edge cases (bank CSV quirks, changing APIs, odd hardware, OAuth failures).
    • Library/framework churn (e.g. webpack4→5, specific Arduino boards).
    • Architecture, long-term maintainability, security, performance, and ops.
  • Pattern described repeatedly: AI makes the first 60–80% trivial, then the remaining 20–40% becomes harder because you’re debugging unfamiliar, often bloated code. Sometimes it’s faster to rewrite by hand.
  • Rapid codegen can overload code review/QA and generate large amounts of technical debt.

Why No Visible Productivity Explosion?

  • Empirical metrics (app stores, some software output graphs) show no obvious inflection; skeptics invoke Amdahl’s Law and Theory of Constraints: speeding up coding (a fraction of the work) doesn’t speed shipping much.
  • Demand, attention, distribution and maintenance remain the real bottlenecks; markets are saturated, and shipping something people want is still the hard part.
  • “AI-generated” is often seen as a quality/liability red flag, so usage is underreported.

Diverging Narratives About the Future

  • Optimists expect exponential capability gains and eventual disruption akin to digital photography.
  • Skeptics see unreliable generation, benchmark games, and a hype bubble more like dot‑com/crypto, with limited real productivity so far.
  • Broad agreement: current tools are powerful assistants, not autonomous app factories.