A year ago the adoption story was one number, the MIT report's 95% of pilots with no return, whose methodology was narrower than its headline, as we set out at the time. The surveys published since are larger, more consistent with each other, and tell a less dramatic story that is more useful to a business deciding what to do next. This briefing reads them together as of the last week of January 2026, with the Irish figures wherever an Irish figure exists.
How much, and where
- 88%
- Organisations using AI in at least one business function, up from 78% a year earlier (McKinsey, 1,993 respondents, 105 countries) [1]
- 20.0%
- EU enterprises with ten or more employees using AI in 2025, up from 13.5% (Eurostat) [2]
- $37bn
- Enterprise spending on generative AI applications and models in 2025, from $11.5bn in 2024 (Menlo Ventures) [3]
| Survey | Fieldwork | Who | Headline |
|---|---|---|---|
| McKinsey State of AI | Jun, Jul 2025 | 1,993 participants, 105 countries | 88% use AI in at least one function; 39% report any EBIT impact; 6% high performers |
| Eurostat | 2025 reference year | EU enterprises, 10+ employees | 20.0% use AI, up from 13.5%; Denmark 42.0%, Finland 37.8% |
| Menlo Ventures | Published Dec 2025 | Enterprise spend analysis | $37bn spend; API share Anthropic 40%, OpenAI 27%, Google 21% |
| Ibec | Jul 2025 | Irish employees | 40% use AI in their work, up from 19%; 81% want more training; 27% have had none |
| Deloitte | Aug, Sep 2025 | 1,854 executives, 14 countries | 1 in 4 Irish firms has a chief AI officer, against 19% elsewhere |
| PwC CEO Survey | Published Jan 2026 | 4,454 chief executives, 95 countries | 17% of Irish CEOs saw AI revenue in the past year, against 29% globally |
| Gartner Hype Cycle | Aug 2025 | Analyst assessment | Generative AI in the trough of disillusionment; agents at the peak |
Two readings survive contact with all of those numbers. The first is that use is now near-universal among large organisations and roughly one in five among all firms above ten staff.[1][2] The second is that use is shallow. McKinsey finds a third of organisations scaling AI across the enterprise and only 6% capturing more than 5% of earnings from it; 62% were experimenting with agents and 23% had scaled one anywhere.[1] The distance between trying a tool and running part of the business on it has not closed, and the surveys are now precise enough to measure it.
| Group | Using at least one AI technology |
|---|---|
| Large enterprises (250+) | 55.0% |
| Medium (50 to 249) | 30.4% |
| Small (10 to 49) | 17.0% |
| Information and communication sector | 62.5% |
| Professional, scientific and technical | 40.4% |
| Most common technology: text mining | 11.8% of all enterprises |
The size gradient is the fact that matters most for an Irish firm of fifty people: a large enterprise is three times as likely to be using AI as a small one, and the gap widened in 2025 as 26 of 27 member states rose.[25] McKinsey's data explains part of why. Among its high performers, 48% of respondents strongly agreed that leaders demonstrated ownership of AI; among everyone else, 16%. Firms with revenue above $5bn reached the scaling phase 48% of the time, smaller firms 29%.[26] The variable is not the tool. It is whether someone senior owns it.
Ireland, specifically
The Central Statistics Office's own count, published in February 2025 for the 2024 reference year, had 15% of Irish enterprises using AI, up from 8% the year before, with 51% of large enterprises doing so; the 2025 figure is due in February.[27] That puts Ireland near the European average and a long way behind the Nordic leaders.[2] The executive surveys add texture. Deloitte's fourteen-country study found one in four Irish firms had appointed a chief AI officer against 19% elsewhere, which suggests governance ahead of delivery.[5] PwC's survey of chief executives, published at Davos in January, found only 17% of Irish respondents had realised additional revenue from AI in the past year against 29% globally, and Irish confidence in revenue growth at its lowest in the survey's history.[6] Ibec's survey of employees found usage had doubled in a year to 40%, that 81% of workers said they could do more with training, and that 27% had received none.[4] Adoption in Ireland is being pulled from below by staff and set from above by governance, with the middle, where a workflow gets changed, still thin.
Where the value showed up
Software development was the runaway case. Coding tools took $4.0bn of the $7.3bn enterprises spent on departmental AI in 2025, 55% of it and more than IT, marketing, customer service, design and human resources combined, and Anthropic's share of that market drove its 40% of API spend, up from 12% two years earlier.[3] Customer service and document-heavy work came next, with smaller and more contested gains. OpenAI reported a 40% average reduction in ticket handling time across its enterprise customers, which is a vendor's number.[8] JPMorgan's internal assistant reached 200,000 daily users who reported saving three to six hours a week on document work.[9] Klarna, the most-cited automation story of 2024, spent 2025 walking part of it back: its chief executive told Bloomberg in May that cost had been too dominant a factor and quality had suffered, and the company began hiring human agents again while the assistant kept most routine conversations.[10][11]
| Company | What | Reported |
|---|---|---|
| JPMorgan Chase | Firm-wide assistant on OpenAI and Anthropic models | 200,000 daily users; 3 to 6 hours a week saved; 450+ use cases in production |
| Klarna | AI customer-service agent, then partial reversal | Two-thirds of chats automated in 2024; human agents rehired from May 2025 after quality fell |
| AIB | Microsoft 365 Copilot | Rolled out to most of about 10,000 staff from July 2025 |
| Shopify | Company-wide mandate | AI use a baseline expectation; teams must show AI cannot do a job before hiring |
| Accenture | Reskilling and AI-linked restructuring | 550,000 staff reskilled; about 11,000 exited in September 2025 as part of an $865m programme |
| Goldman Sachs | GS AI Assistant on OpenAI, Google and Anthropic models | 10,000 staff in January 2025, firm-wide from 23 June |
| Salesforce | Agentforce, sold to customers and used on its own support desk | $500m standalone annual recurring revenue by December; support headcount cut from about 9,000 to about 5,000 |
| IBM | AskHR chatbot | 94% of HR interactions handled without a person, 11.5m in 2024; HR net promoter score from −35 to +74 |
| Amazon | Corporate restructuring | 14,000 corporate roles cut in October, the chief executive citing generative AI and agents |
Against the vendor case studies stand the two controlled experiments of the period, and they point in opposite directions. Brynjolfsson, Li and Raymond's study of 5,179 customer-support agents, published in the Quarterly Journal of Economics in 2025, found a generative-AI assistant raised issues resolved per hour by 14% on average and 34% for the least experienced workers, with almost no effect on the most experienced: the tool diffused the top performers' habits downward.[32] METR's trial of sixteen experienced open-source developers on their own repositories, published in July, found the opposite at the top of the skill range: with AI allowed, they took 19% longer, having predicted a 24% speed-up and still believing afterwards that they had been 20% faster.[33] Read together, the experiments say the gain is largest where the work is routine and the worker is new, and may be negative where the work is complex and the worker is expert. That is a more useful map than any adoption percentage.
The workforce
There was no evidence of economy-wide displacement in 2025, and clear evidence of an effect at the entry level. Stanford's study of payroll records from ADP, published in August, found employment of 22 to 25 year olds in the most AI-exposed occupations about 13% below where it would otherwise be since late 2022, while older workers in the same occupations were unaffected; the effect ran through reduced hiring rather than redundancies.[15] Employers cited AI for 54,836 announced US job cuts in 2025, about 5% of the total.[16] The company-level stories were mixed: Accenture tied cuts to staff who could not be reskilled while reporting half a million reskilled, and Shopify's chief executive made AI use a condition of asking for headcount.[13][14] In Ireland the constraint ran the other way, with four in five workers asking for more training than they had had.[4]
Why the pilots still died
The MIT figure measured the absence of a financial baseline in a non-random sample, and it has held up as a description of a management problem rather than a technology one; its own explanation was that tools which do not learn from feedback stall after the demo, and that purchased tools reached production about twice as often as internal builds.[17] Gartner's placement of generative AI in the trough of disillusionment in August was the same finding from the analyst side: pilots that impress and do not scale.[7] The shadow numbers explain part of it. In the MIT sample 90% of workers used personal AI tools while 40% of employers had bought any; the adoption was happening from the bottom, on personal accounts, outside any policy.[17]