Will AI productivity make everyday life more affordable?

Scope

Australian household affordability over the coming decade

Confidence

Moderate on possible productivity gains; low on household price effects

This is a provisional assessment of an open question. It distinguishes the available evidence from Alex’s established submitted positions. It does not attribute a settled new policy position to him.

Follow the whole chain

A 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.

The 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.

Evidence for a substantial opportunity

The Productivity Commission’s final December 2025 report, 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.

That 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.

Evidence against a simple shortcut

METR’s early-2025 randomised experiment 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 points towards speedups but reports serious selection and measurement problems. Neither result is a universal current productivity estimate.

These 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.

What to measure next

For 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.

The 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.

What would change this answer

Credible 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.

Limits of this edition

No 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.