What the finding establishes

Self-employment and non-employing businesses are a large, persistent part of Australia's economy. Software AI can plausibly lower some capability barriers for a person or microbusiness. Current official data do not isolate how much Australian business formation is caused by AI, who benefits, or whether the work is viable. Creation is therefore a material pathway to investigate, not the established dominant labour-market effect.

HighSelf-employment and non-employing businesses are economically material.
MediumSoftware AI plausibly lowers some capability barriers.
LowThe current Australian magnitude and distribution of AI-caused creation.

Evidence in the threshold test

A large existing pathway

Australia had 2,729,648 actively trading businesses at 30 June 2025. The government-derived count used by the report records 1,735,470 non-employing businesses. This establishes material scale, not AI causation or durable income.

An all-age population

Owner-managers are not only young workers. The 2021 Census records owner-managers as 13.8% of employed Australians, with age and gender patterns that make distribution part of the test.

Software and robotics differ

A person can rent software capability using ordinary computing equipment. Robotics normally requires equipment, premises, integration and maintenance. The lower-capital creation mechanism transfers poorly to many physical roles.

Transition signals are uneven

Australian official analysis found no clear current decline in entry-level roles due to generative AI and no current evidence of widespread displacement. Overseas and robotics evidence identify different risks and populations.

What this verdict does not mean

  • creation is larger than displacement or augmentation
  • new businesses survive or provide good work
  • active support is sufficient to improve outcomes
  • robotics offers the same pathway as software
  • one of the series pathways is more likely than another

Where this argument could be wrong

The threshold verdict would weaken if linked data showed AI-enabled formation is negligible, short-lived or confined to people already well placed to succeed. It would strengthen if Australian longitudinal data linked AI use to survival, earnings and entry/re-entry outcomes across age, gender, region, disability and prior occupation.

Bottom line

AI-enabled creation belongs in the labour-market account, but the evidence does not yet show it wins. Measuring only jobs lost misses work people create; counting businesses created misses income, survival and access.

Selected source notes

These links support the bounded public extract. Source inclusion does not imply endorsement of the report's synthesis.

Suggested citation

Molony, Rick (2026). “Is new job creation real enough to matter?” Report 1 extract, AI Policy, Our Resilient World. CC BY 4.0.