Ireland's AI dividend is arriving before its training bargain

University College Dublin's release puts a useful denominator under workplace AI in Ireland. The Working in Ireland Survey 2025 covered 4,300 workers across the Republic of Ireland and Northern Ireland. Almost 30% reported using generative AI at work. That is substantial adoption, but the aggregate conceals a steep access gradient.

In the Republic, reported use rose from 8.5% among workers earning less than €15,000 net a year to 73.8% among those earning €110,000 or more. People with postgraduate qualifications were up to 15 times more likely to use AI than workers with basic qualifications. Professional and managerial employees were five to six times more likely to use it than people in caring, trades, process or machine roles. Large firms were more than 1.6 times as likely as small firms to employ AI users.

Those comparisons do not show that income or education causes adoption. They do show why a company-wide usage rate is an inadequate skills metric. Access to suitable tasks, licensed tools, data, managerial permission and time to learn may all sit behind the gaps. A workforce plan needs to identify those mechanisms rather than label non-users as resistant.

Training is lagging behind use

Employer-provided generative-AI training reached 42.8% of employees in the Republic and 37.7% in Northern Ireland. Among trained workers in the Republic, almost 60% reported less than a full day of instruction. Only around half of employees in the Republic, and just over a third in Northern Ireland, worked for organisations with an official AI-use policy.

That combination matters. Self-teaching can spread useful practice quickly, but it leaves workers to infer where confidential data may go, which outputs require verification and when not to delegate. A one-off awareness session also cannot substitute for supervised practice in a real workflow. Training should therefore be measured by demonstrated task performance and escalation behaviour, not attendance.

The distribution of benefits is equally unsettled. One third of AI users said their work pace had intensified. Only 4.7% reported higher earnings associated with AI use. Among those reporting time savings, many redirected the time into more work; some took on routine tasks and others shifted toward more complex or creative work. RTÉ's independent report retained that ambiguity rather than treating every saved minute as a worker benefit.

The evidence is a map, not a causal verdict

The release identifies Ipsos B&A as the fieldwork provider and gives the fieldwork dates as 15 May to 28 August 2025. It links to the full report, but the summary itself does not reproduce questionnaire wording, weighting, response rates or confidence intervals. The reported outcomes are self-reported. Workers who already have more autonomy and digital support may be more likely both to use AI and to report benefits. The survey therefore cannot establish that AI caused higher work intensity, wage outcomes or differences between groups.

A broader 35-country European study, using more than 36,600 workers from the 2024 European Working Conditions Survey, also found adoption concentrated among skilled and cognitively non-routine jobs. Its early shift-share analysis found no detectable technology-related task restructuring. That is not a contradiction: the studies use different periods, measures and designs. It is a warning against converting adoption correlations into a productivity or displacement claim.

For decision-makers, the immediate move is to build an AI access-and-return ledger by occupation. Record who has an approved tool, what task it supports, hours of supervised practice, quality checks, time saved, workload change and any pay or progression outcome. Split results by employment status, location, income band and employer size.

Then treat capability as a work-design problem. Give lower-access groups protected learning time, task-specific examples and a route to challenge bad outputs. Agree in advance how verified savings will be used: reduced backlog, better service, learning time, shorter hours or shared financial gain. Without that bargain, adoption can rise while trust and opportunity narrow. The Skills Atlas can structure the capability categories; the organisation still has to measure access and distribution.