On 24 August, the Expert Group on Future Skills Needs published its note on the Irish labour market and AI, and the coverage told us we were a global front runner. Second in Europe. Generating AI jobs at two to three times the pace of our neighbours. The Minister called the pace of change astounding.
So I read the note. Then I used AI to track down the sources the note is built on. Nothing in it is dishonest. But nothing in it tested a single person either, and the gap between what these numbers measure and what they were used to say is wide enough to drive a press release through.
There are five numbers doing the work. Four of them are thin.
The first is the jobs figure, and it comes from Indeed. It counts the share of job adverts containing certain words. Machine learning. Data science. Artificial intelligence. ChatGPT. That is a measure of how recruiters write copy, not of what work exists or who can do it. Indeed's own researchers went and read the text around those mentions and found that roughly a quarter of them had no real context at all, which is to say a quarter of our AI boom is a bullet point somebody added to a job spec. It is also a share, not a count, so when overall hiring falls the AI percentage rises without a single new job being created. The note itself reports that ICT hiring here has softened. Both things cannot be evidence of the same trend.
The second is the talent figure, and it is the one I would frame and hang on a wall. The OECD counts AI talent by looking at LinkedIn profiles, dividing the number of members who list AI skills or hold an AI job title by the total number of LinkedIn members in the country. And LinkedIn's definition of an AI skill includes AI literacy skills such as ChatGPT and GitHub Copilot. So you can type ChatGPT into your profile at lunchtime and be counted as part of Ireland's AI talent base by dinner. The OECD attaches a health warning to this data, saying plainly that the figures are influenced by how many people in a country use LinkedIn and should be treated with caution. That warning is in the source. It is not in the headline. Somewhere between the database and the podium it quietly fell off the back of the lorry.
The third is diffusion, where Ireland is placed first in the EU. That ranking traces back to Microsoft telemetry, a measure of how many working-age people used a generative AI product, derived from Microsoft's own product data. Consider what that measures in a country whose office parks are wall-to-wall Microsoft shops, where corporate 365 licences are near universal and Copilot arrives bundled inside them. We are not leading Europe in AI adoption. We are leading Europe in having a lot of Microsoft licences, which is not the same achievement, however much it looks like one from a distance.
The fourth is individual usage, where 45% of us are said to use generative AI against an EU average of 34%. The threshold for that is one use in three months. One. And 2025 was the first year Eurostat asked the question, so there is no previous year to compare it against, which does not stop the note describing it as rising rapidly. The note also concedes, in the same breath, that most of that use is private rather than work or study. Someone asking ChatGPT for a lasagne recipe in June is, statistically, part of Ireland's AI transformation.
Then there is the fifth number, and this is where it gets interesting, because the fifth number is good. It comes from the Eurostat and CSO enterprise survey, a proper representative sample of Irish businesses, asked a consistent question, the same way, in every country in Europe, under EU regulation. It is the only instrument in the entire exercise that actually went and asked anybody anything.
In 2025, 20.2% of Irish enterprises with ten or more employees used AI. The EU average was 20.0%. Denmark was on 42%. Finland 38%. Sweden 35%.
Ireland is exactly average. Not second, not a front runner. Average.
That number is in the report. It appears as Figure 7. It is broken out by company size, to make a genuinely important point about large firms racing ahead while small ones stall, and it is never once ranked against the rest of Europe. The single most reliable measurement we have of whether Irish businesses actually use this technology went into the document and never came out the other side.
I don't think that's a conspiracy. I think it is what happens when four flattering indicators and one honest one arrive on the same desk.
And here is what actually bothers me about it, because it isn't really about AI at all. Almost every favourable ranking in that note is a measurement of who is located here rather than what we can do. Multinational headcount. ICT specialists on foreign payrolls. Inward migration of skilled workers. Microsoft product penetration. Notice who else keeps appearing at the top of these tables alongside us. Luxembourg. Two small open economies with enormous foreign corporate sectors and small denominators. That is not a coincidence, it's a pattern, and I've written before about how much of our national story is really a story about hosting other people's companies.
We are measuring our address and calling it an achievement.
In fairness to the note's author, Diarmaid Smyth, he does not overclaim once. He says plainly that it is too early to be definitive about AI and job losses. He points out that the softening in tech may be a post-pandemic rebalancing rather than an AI story. He flags research suggesting hybrid working, not AI, may explain weak junior hiring. He argues for focusing on tasks rather than jobs, which is right. And his strongest section makes the case that AI skills are far broader than AI specialists, that what matters is managerial capability, sector-specific application, data literacy, governance and responsible use, spread across every industry rather than concentrated in a technical elite. That argument deserved the headline. It didn't get it.
The report is considerably more careful than the coverage it produced. The word astounding does not appear anywhere in it.
So what would tell us something real?
Lead with the enterprise survey. It is the best instrument we have and it says we are average, which is a far more useful place to start a skills policy than a lap of honour. Being average is not a scandal. Being average and believing you are second is.
Measure capability instead of exposure. Every indicator in that note counts activity. Mentions, profiles, sessions, one use in three months. Not one of them asks whether a person can competently do anything. We would not accept this anywhere else in our lives. Nobody gets a driving licence for confirming they have been in a car. Revolut verifies three million Irish identities because money matters enough to check. Skills apparently do not.
And carry the caveats forward. The OECD warns about its own data. Eurostat says the word assumed out loud. Those qualifications exist at the source and are stripped away at every retelling until, by the time they reach a Cabinet briefing, four proxies have hardened into a fact.
None of this means Ireland is doing badly at AI. I suspect we are doing reasonably well, and I have a business that depends on us doing better. It means we don't know, in any rigorous sense, and the instruments we are using are structurally incapable of telling us.
Disclosure: I run a company that sells AI education and certification (and you better believe I used AI to help create this article) so I have an obvious interest in arguing that assessment beats self-report. Read the above accordingly, and check the sources yourself. They are open, free and linked throughout.
