Time saved by AI offset by new work created, study suggests

Original Article ↗ Hacker News Discussion ↗

Overall view of AI and work

  • Many see AI fitting a long pattern: new technology increases productivity but doesn’t reduce total work; it shifts what work is done (Jevons paradox, “backlogs never shrink”).
  • AI is compared to PCs, washing machines, tractors, the Internet, and search: each removed some tasks but led to more output, more data, and new jobs, not mass leisure.
  • Several argue the real constraint is capitalism and power: gains go to owners, not shorter hours; without strong labor policy, efficiency just raises expectations.

What the study actually shows

  • Multiple commenters highlight the key result: average time savings of ~2.8% (about 1 hour/week) and essentially zero measured impact on earnings, hours, or wages.
  • Some say the article’s “time saved offset by new work” framing is misleading: the paper shows small benefits, not clear offsetting by verification work.
  • Others note the data is from late 2023 and only covers people still employed, so may understate future impact.

Developers’ real‑world experience (highly mixed)

  • Enthusiasts report 2x–10x speedups on certain tasks: boilerplate, small features, scrapers, Terraform, Go/Python/JS glue code, tests, and “blank page” problems.
  • Many others see modest or negative net gains: AI often emits wrong, outdated, or invented APIs; they spend extra time prompting, debugging, and reviewing “vibe‑coded” blobs.
  • Utility is very language/domain dependent: better for Python/JS and simple infra; worse for Swift/iOS, complex multi‑service backends, and systems programming.
  • New work appears in prompt engineering, test creation, code review, and maintaining AI‑generated code; some feel they’re now “AI response mechanics.”

Quality, determinism, and “toolness”

  • Several complain about AI‑inflated corporate emails and support replies: longer, vaguer, harder to parse.
  • Some insist AI isn’t a deterministic tool and call it a “parlor trick” or “occult divination”; others respond that non‑determinism doesn’t preclude productivity, and local models can be deterministic.
  • There’s strong skepticism of claims like “100x productivity”; people demand metrics and note that architecture, communication, and debugging still dominate dev time.

Distributional and social effects

  • Concerns that AI erodes junior and low‑skill roles (clerks, customer service, retail, warehouse) while creating fewer, harder “boundary” jobs.
  • Fear that skills in “using AI” become basic literacy, further commoditizing mid‑skill labor and pushing more people into precarity or “bullshit jobs.”
  • Debates over AGI and post‑scarcity: some imagine abundance and UBI; others expect entrenched power, tighter control over resources, and persistent inequality.