The federal government's first proper study of AI and employment landed last month, and one number in it should have settled an argument.
Employment in software development has not fallen in Australia.
It has grown 25 per cent since November 2022, to around 199,000 people, which is 40,000 more than in the quarter before ChatGPT arrived.
The report goes further and says the much-discussed difficult labour market for software developers is not evident in the aggregate Australian data.
But, but….
If you have spoken to anyone trying to get a job in tech lately, that will not match a word of what you have heard.
Online forums are full of people who have sent out hundreds of applications and heard almost nothing back.
Both things are true at the same time, and working out why is more useful to a business owner than either statistic on its own.
Part of the answer is that employment counts people who already have a job, not people trying to get one.
A workforce can keep growing while the door into it narrows. The squeeze is sharpest at entry level.
Research from Stanford's Digital Economy Lab, built on payroll records rather than surveys, found employment for software developers aged 22 to 25 in the United States fell almost 20 per cent from its late-2022 peak.
Australia has not shown that pattern in the aggregate yet, though one local study has picked up lower hiring rates for younger workers in exposed occupations.
The rest of the answer I ran into myself. We advertised for junior developers recently and had no shortage of applicants.
What we struggled to find was anyone who could show us what they had actually built with AI.
Nearly every candidate said they loved AI and used it every day.
Very few could point to a single thing they had made with it. No AI-integrated product, no site they had put together, nothing shipped.
What arrived instead were the same resume templates, the same lists of previous roles and the same cover letters, plenty of which had clearly been written with AI without ever demonstrating anything built with it.
The bottom rung
That gap makes more sense once you think about what AI has absorbed.
The work juniors traditionally cut their teeth on – the boilerplate, the basic tasks, the first pass at testing – is precisely what these tools handle well.
The bottom rung of the ladder has been sawn off.
What employers need at that level now is someone who can direct the tools and produce something real with them, which is a different skill from writing code line by line.
So, the requirement has shifted while the signals candidates send have stayed exactly where they were.
A portfolio has never been cheaper or faster to build than it is right now, which is why turning up without one says more than it used to.
Anyone can spend a weekend shipping something with these tools.
The person who has done it three times is telling me something a tidy resume never could.
Where's the proof?
If you are hiring, the practical change is to stop screening on the old proxies and start asking for evidence.
Ask what someone has built, not what they have used. Ask to see it running.
In my experience that separates candidates faster than any interview question about their favourite model, and it surfaces capable people whose resumes would never have made the shortlist.
If you are on the other side of this and trying to get in, the same logic runs in reverse.
Another certificate will not move you up the pile. A handful of things you have built and can show, ideally solving a problem somebody actually had, will.
That is the cheapest advantage available to a job seeker right now, and it is sitting there largely unclaimed.
It would be too neat to pin all of this on AI.
What employers want
A good deal of the entry-level slowdown is ordinary economics, with budget cuts and interest rates working their way through the tech sector, and the aggregate growth does hide a real junior squeeze.
But the part employers can do something about today is what they choose to screen for.
At Smec AI, the federally funded AI adoption centre I lead, we spend our days helping businesses put AI to work.
The same test we apply to them turns out to apply to the people we hire, which is simply whether anything has actually been built.
Everyone in the interview says they love AI.
Almost nobody can show me what they made with it.
