The AI Skills Bluff

I sat in on an interview a few months back, dialled in as a favour to a hiring manager who wanted a second opinion. The candidate’s title said AI Automation Lead. His resume said he had built and deployed multiple LLM based agent workflows. Forty minutes in, someone asked him to explain, in plain terms, what happens when a retrieval augmented generation system gets a question it has no good context for. He talked about the model doing the reasoning for him for two full minutes and said nothing that meant anything.

Nobody in the room called it. Everyone wanted him to be real, because they badly needed someone to be real, so they let the fog stand in for the answer.

Here is the thesis, plainly: Australian employers are chasing an AI skill so scarce and so well paid that plenty of candidates have worked out the fastest route to having it is describing it convincingly. If you are hiring an AI engineer, a data lead, or a prompt literate digital marketer in Sydney or Melbourne this year, the useful question is not do they know AI. It is how would I actually find out, in the room, before I sign the offer. The plain answer is that you stop grading vocabulary and start asking about failure, which is the one thing a rehearsed answer cannot survive. That holds whether you are filling a six figure AI engineering role or a mid level marketing seat that now expects fluency with the same tools.

Because the money is real, even where the resume is not. PwC’s 2026 AI Jobs Barometer found AI skilled workers in Australia now command an average wage premium of 62 per cent, up from 57 per cent a year ago. In technology, media and telecommunications specifically, the premium sits around 59 per cent. AI related job postings across all sectors jumped from roughly 20,000 in 2024 to 41,000 in 2025. That is not a gentle trend line. That is a market throwing money at a skill it cannot verify quickly, which is exactly the condition under which people learn to perform competence rather than build it.

Set that against the rest of the economy and the gap gets stranger still. The ABS Wage Price Index for the March quarter 2026 showed wages nationally rose 3.3 per cent over the year, a touch softer than the 3.4 per cent recorded twelve months earlier. Ordinary wage growth is cooling gently. The AI premium is doing the opposite, and doing it hard. When one narrow slice of the market pays fifty or sixty points above the going rate, expect the gold rush behaviour that comes with it to show up uninvited in your interview pipeline.

The broader labour market, meanwhile, is not roaring either. ABS Labour Force figures for May 2026 put the unemployment rate at 4.4 per cent, with employment up by around 40,000 for the month and the number of unemployed people falling by roughly 18,000, a market that reads as steady rather than tight. The Reserve Bank held the cash rate at 4.35 per cent at its June meeting, noting the unemployment rate had actually come in higher than expected in April even as other measures of the labour market held up reasonably well. Put plainly, the general jobs market is cautious and a little uneven. The AI segment sitting inside it is overheating on its own separate timetable. That mismatch, a soft broad market next to a scorching narrow one, is precisely why the bluff works as often as it does. Hiring managers are anxious about filling a tight, high stakes role, and anxious hiring managers ask softer questions.

Stop asking about tools and start asking about failure. Anyone can say they use Claude, Copilot, or a RAG pipeline in a job interview. Far fewer people can tell you, unprompted, about the time it went wrong, what broke, and what they changed because of it. Genuine practitioners carry scar tissue. They will mention the hallucination that embarrassed them in front of a client, the prompt that worked beautifully in testing and fell over the day it hit production, or the workflow they abandoned outright because the unit economics never stacked up. Bluffers only have the highlight reel, because the highlight reel is all they built.

Ask for the boring middle, not the impressive headline. Walk me through the thing you tried that did not work filters harder than any certificate on a resume. Ask what they would build differently with half the budget and half the time. Ask what they still would not trust AI to do unsupervised, in their own domain, right now. A candidate who cannot name a genuine limitation is not being modest. They have simply not gone deep enough to hit one yet, and that is useful information in itself.

Then look at your own panel honestly. If nobody interviewing has hands on experience shipping something with these tools, you are grading an essay written in a language you do not speak. Bring in one person who has actually built the thing, even for twenty minutes on the call, before the offer goes out. It is a smaller ask than it sounds and it will save you a much larger conversation in month three, the one where the new hire’s automation pipeline turns out to be three scheduled prompts and a spreadsheet.

This is not only a tech hiring problem. Digital marketing teams are living through the same pressure test, because AI led content strategy has become nearly as common a resume line as results driven once was. The test holds regardless of the function: ask for the failure, not the feature, and the bluff usually surfaces before the second coffee.

Tell the truth about depth and let evidence carry the argument. If you have used a tool for three months on a single project, say exactly that, then talk properly about the one thing you learned rather than gesturing at a breadth you do not actually have. Sydney’s sharper hiring managers, the ones worth working for in the first place, are now explicitly screening for people who can explain trade offs and limits over people who can recite keywords fluently. A vague capability claim is a bigger risk to your candidacy in 2026 than a narrower, demonstrated skill set, because AI assisted screening tools have become noticeably better at spotting a resume that names a tool without ever showing what it actually produced.

The M.C. Escher line applies here neatly, even though he was thinking about lithographs rather than LinkedIn profiles:

We adore chaos because we love to produce order.

One. Replace tell me about your AI experience with tell me about the AI project that did not work, and why. The second question is close to impossible to fake convincingly and it takes roughly ninety seconds to expose a bluff that a resume hid for weeks.

Two. Weigh the AI premium against your baseline before you agree to pay it. If economy wide wages are growing at 3.3 per cent and you are being asked to stretch fifty to sixty per cent further for an AI skilled hire, treat that premium as a claim requiring proof, not a market rate you accept on faith. Ask for a work sample, a repository link, or a short live task before you sign off on the number.

Three. If you are the candidate, build one piece of evidence you can point to instead of a paragraph of adjectives. A single documented project, including exactly where it broke and how you fixed it, will outperform five bullet points listing tools you have merely opened once.

This week, before your next AI focused interview, write down one question that forces the candidate to describe a failure rather than a feature. Ask it early, and watch what happens to the room.

I have been recruiting technology and digital talent since 1998, through enough booms and busts to recognise a gold rush when I am standing knee deep in one, and this is among the fastest and messiest I have seen candidates and employers try to read each other honestly. At Big Wave Digital we have started building failure based questions into our own screening process for exactly this reason, because a confident answer is never automatically the same thing as a correct one.

This newsletter is free and will stay that way. If it was useful, forward it to the hiring manager in your life who just described their new hire as basically a prompt engineering genius.