Has the cost of building software dropped 90%?

Original Article ↗ Hacker News Discussion ↗

Headline and Evidence

  • Many commenters reject the “90%” claim outright, invoking Betteridge’s law and criticizing the article’s unit-less graph and lack of empirical support.
  • Several point out that any “cost drop” ignores massive GPU/datacenter spending and current unprofitable AI economics.
  • Multiple people ask: if costs really dropped that much, where is the observable explosion of high‑quality, cheap software?

Reported Productivity Changes

  • Experiences are polarized. Some report modest gains (e.g. 30–50% faster on coding tasks) or even being slower with AI; others claim 5–10x speedups for solo dev work, quick feature shipping, prototypes, internal tools, and personal utilities.
  • Consensus that AI is most useful for:
    • Boilerplate, scaffolding, simple CRUD, integrations, scripts.
    • Navigating and understanding codebases, debugging, and refactors.
  • It struggles with complex, messy, long‑lived systems, subtle logic, and maintaining behavior without regressions.

Quality, Testing, and Maintenance

  • Strong skepticism about “300 tests in a few hours”: many say AI-written tests are often superficial, redundant, or outright wrong and require heavy review.
  • Several note that human-written production code is often terrible too, so AI “slop vs human slop” is not obviously worse—but AI amplifies code volume, which increases long‑term maintenance cost.
  • A repeated theme: building v1 may be cheaper, but maintenance, feature evolution, debugging, security, and organizational risk still dominate total cost.

Career, Skills, and Developer Anxiety

  • Many developers are anxious about “how to position” themselves. Common advice:
    • Deepen domain/business knowledge and move from “specs-to-code” to “solve business problems”.
    • Broaden to full‑stack, product, or PM‑adjacent roles, or specialize in hard/low‑level areas less amenable to automation.
    • Treat LLMs as powerful junior partners: learn prompt design, agent orchestration, and project management of AI.

SaaS, Internal Tools, and Spreadsheets

  • Contrary to the 90% thesis, several note no visible collapse of major SaaS players or tidal wave of new SaaS, though some report:
    • Companies replacing expensive SaaS (e.g. ETL, Salesforce‑like tools) with cheaper in‑house systems now feasible with AI.
    • Solo/indie devs targeting small niches that previously weren’t worth building for.
  • Big debate over replacing “core” spreadsheets: spreadsheets are flexible and empower domain experts but become opaque, error‑prone “shadow IT.” Some see AI‑built internal apps (Streamlit, Rails, etc.) as a partial upgrade; others argue most such replacements just reinvent Excel badly.

Organizational and Hype Constraints

  • Many stress that coding is only a fraction of software cost; coordination, requirements, change management, and support dominate, and AI doesn’t fix that.
  • Several compare current claims to self‑driving cars and past “software is dead” moments (outsourcing, low‑code).
  • Overall sentiment: AI coding tools are genuinely useful and sometimes transformative at the margin, but “90% cheaper software” is not yet visible in real organizations.