AI adoption and Solow's productivity paradox

Original Article ↗ Hacker News Discussion ↗

Enterprise AI tools & uneven adoption

  • Many commenters argue that large organizations “use AI” only superficially: buying Microsoft Copilot or similar add‑ons, then discovering they can’t see spreadsheets, emails, or internal systems in useful ways.
  • Risk, security, and IP concerns lead to highly constrained deployments, so AI remains a chatbot in a sidebar rather than an integrated actor in workflows.
  • Some companies are pushing usage quotas (“use more tokens”) without clear value propositions, causing employees to generate pointless traffic or “internal fan fiction” just to hit KPIs.

Developers vs other white‑collar workers

  • Software engineers report strong gains from coding agents: faster scaffolding, PR generation, refactors, log analysis, and working in unfamiliar stacks. Solo or small‑team developers see the clearest net benefit.
  • Others say the speedup is limited by non‑coding work (requirements, reviews, coordination), so overall velocity barely changes.
  • Outside engineering, LLMs feel more like “intelligent autocomplete”: small wins for slide decks, emails, transcription, and lookup, but rarely transformative.

Verification, ‘slop’, and organizational bottlenecks

  • A recurring theme is “AI slop”: large, low‑quality PRs, verbose reports, and long documents that someone must still review.
  • Several point out that 98% automation is useless if the 2% of errors can’t be reliably detected; then everything needs human review and much benefit evaporates.
  • Open source maintainers and internal reviewers report fatigue from AI‑generated patches and code reviews that are technically valid but noisy, superficial, or misaligned with project standards.
  • Many argue that large firms are I/O‑bound, not CPU‑bound: meetings, approvals, and inter‑team dependencies dominate timelines, so faster code or text generation doesn’t move the system much.

Productivity paradox & economic impact

  • The thread frequently references the “productivity paradox”: like early IT, AI may require years of experimentation, re‑organization, and capital outlay before aggregate productivity shows up in statistics.
  • Others are skeptical that LLMs belong in the same category as computers or electrification, seeing them mainly as text generators whose costs (GPU capex, energy, hype‑driven waste) currently outweigh measured gains.

Future of work & structural change

  • Some foresee fewer engineers per company but far more small companies: one expert plus agents replacing larger teams, especially for niche products.
  • Others predict a “hollowing out”: junior developers and many white‑collar roles displaced or deskilled, with long‑term maintainability and domain knowledge becoming scarcer.
  • There’s broad agreement that autonomous, trusted agents (not just chatbots) are the missing piece for large, visible productivity jumps—but also the hardest to deploy safely.