{
  "slug": "productivity-and-affordability",
  "title": "Will AI productivity make everyday life more affordable?",
  "topic": "Living standards",
  "answer": "It could help, but a faster task is several steps away from a cheaper household bill. The size, timing and distribution of gains depend on implementation, competition, wages, prices and supply constraints.",
  "scope": "Australian household affordability over the coming decade",
  "standing": "Open question · provisional evidence synthesis",
  "confidence": "Moderate on possible productivity gains; low on household price effects",
  "body": "## Follow the whole chain\nA productivity claim needs a unit: tasks per hour, quality-adjusted output, firm value added or national output. Affordability asks a different question about what households can buy with their incomes. A saved hour does not itself reveal whether prices fall, wages rise, profits increase or more work is attempted.\n\nThe question is open in this series. Alex’s submitted position supports useful AI capability and substantial justified infrastructure investment, while opposing general subsidies for AI consumption. It does not promise that productivity will solve the cost of living. [Commonwealth submission, 30/09/2026](/submissions/commonwealth/).\n\n## Evidence for a substantial opportunity\n[The Productivity Commission’s final December 2025 report](https://assets.pc.gov.au/2025-12/data-digital_0.pdf), printed page 12 and Appendix B.2, considers multifactor-productivity gains above 2.3% over a decade likely, with considerable uncertainty. Its calculation combines 14% affected activity with 16.4% gains within affected activity. The associated 4.3% labour-productivity uplift maps to about $116 billion of GDP at 2023–24 prices, holding population, hours and employment fixed. This is a conditional GDP-level comparison, not accumulated savings or money already delivered to households.\n\nThat projection gives a reason to take the opportunity seriously. It does not supply evidence of price pass-through in housing, food, energy or care. Those sectors have different cost structures and constraints.\n\n## Evidence against a simple shortcut\n[METR’s early-2025 randomised experiment](https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/) found a 19% increase in completion time across 246 tasks by 16 experienced developers. It concerns particular early tools and familiar repositories. Its [February 2026 follow-up](https://metr.org/blog/2026-02-24-uplift-update/) points towards speedups but reports serious selection and measurement problems. Neither result is a universal current productivity estimate.\n\nThese studies also illustrate why perceived speed, measured time and economic value need separate treatment. Producing more material may create review work. A higher-quality service may be valuable even if its sticker price does not fall. A firm may use savings to expand output rather than reduce prices. These are possible channels, not measured outcomes asserted for Australia.\n\n## What to measure next\nFor households, an informative assessment would track real disposable income, relevant prices and service access, split by income and location. For firms, it would include software fees, integration, supervision, errors and ongoing maintenance alongside output. The same saving should not be counted once as lower business cost and again as a separate benefit without checking how the accounts connect.\n\nThe distribution matters. An economy can become more productive while a displaced worker loses income. Infrastructure can create a broad benefit while imposing uneven local costs. The series should make those trade-offs visible rather than treat aggregate output as a complete welfare measure.\n\n## What would change this answer\nCredible Australian evidence connecting actual AI deployment to quality-adjusted output, consumer prices and household income would strengthen or narrow the affordability case. Gains confined to a small set of firms would imply a different answer from widespread competitive pass-through. Persistent implementation costs or harms exceeding the gains would weaken the case.\n\n## Limits of this edition\nNo household saving is forecast here. The Commission’s scenario is not an estimate of the full net welfare effect of AI. It cannot be used to promise a particular reduction in anyone’s bills.",
  "sources": [
    "pc",
    "metr25",
    "metr26"
  ],
  "published": "2026-10-18",
  "evidence_through": "2026-10-18",
  "version": "v1",
  "release": "2026-10-18-v1",
  "change": {
    "old": "No earlier site edition.",
    "new": "Initial publication of this question and its evidence assessment.",
    "why": "Launch of the living-article series, grounded in the dated submissions and primary sources listed here.",
    "confidence": "Moderate on possible productivity gains; low on household price effects",
    "practical_effect": "Establishes a transparent baseline for later review. No earlier public revision is claimed."
  },
  "source_records": [
    {
      "id": "pc",
      "title": "Harnessing data and digital technology: final inquiry report",
      "author": "Productivity Commission",
      "published": "2025-12-19",
      "url": "https://assets.pc.gov.au/2025-12/data-digital_0.pdf",
      "kind": "Conditional economic projection",
      "limitations": "Printed page 12 and Appendix B.2, pages 142–146. A 2.3% multifactor-productivity calculation combines 14% affected activity and 16.4% gains within that activity. The corresponding 4.3% labour-productivity uplift maps to about $116 billion GDP at 2023–24 prices, holding population, hours and employment fixed. This is a conditional level comparison, not cumulative cash, observed savings, price pass-through or net welfare.",
      "checked": "2026-10-18"
    },
    {
      "id": "metr25",
      "title": "Measuring the impact of early-2025 AI on experienced open-source developer productivity",
      "author": "METR",
      "published": "2025-07-10",
      "url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/",
      "kind": "Randomised task experiment",
      "limitations": "16 developers and 246 tasks on familiar open-source repositories; early-2025 tools increased completion time by 19%. Specific tools, developers and tasks. Pair with the February 2026 update rather than treating this as the current general effect of AI.",
      "checked": "2026-10-18"
    },
    {
      "id": "metr26",
      "title": "We are changing our developer productivity experiment design",
      "author": "METR",
      "published": "2026-02-24",
      "url": "https://metr.org/blog/2026-02-24-uplift-update/",
      "kind": "Follow-up experiment with selection problems",
      "limitations": "Later point estimates suggest speedups but intervals include no effect. Developer and task selection, changed pay and concurrent-agent time measurement make the magnitude unreliable. The authors consider the data weak evidence about the size of change.",
      "checked": "2026-10-18"
    }
  ]
}