Sari la:
At the beginning of 2024, Kristalina Georgieva, the IMF’s Managing Director, wrote something that few leaders took seriously at the time: “Almost 40% of jobs globally are exposed to AI. This technology will deepen inequalities between countries if governments do not act now.” More than two years later, there are several research studies and articles that appear to confirm the warning. And Romania has barely acted over these 24 months when it comes to AI.
This piece does not propose forecasts dedicated to an abstract future of work or science-fiction scenarios involving AI. It does, however, propose a manifesto addressing a tension that is growing steadily, in call centers in Cluj, in multinationals in Bucharest, in boardrooms across the country, and in the villages and small towns that, for various reasons, have been placed at the periphery of the reindustrialisation and digitalisation trend. Alexandru Dănescu and Roland Kristo have written about this tension in a report published last year for Edge Institute.
This is a continuation of the call made a few weeks ago, when we highlighted that there is no institution monitoring this phenomenon, measuring its effects, or managing it. There is simply no one responsible for this topic in the Romanian state’s “organisation chart.”
Before we get into detail, a necessary methodological note: what follows is not speculation about what AI might do in the future. It is a synthesis of what the most rigorous global economic research institutions, the IMF, McKinsey Global Institute, World Economic Forum, and peer-reviewed researchers from Europe’s top universities, are already saying about what is happening now and what lies ahead in the 2026–2030 horizon. Romania does not appear in these analyses as a lucky exception. When it appears at all, it appears as a case of maximum risk.
The Double Trap: Exposed and Slow at the Same Time
McKinsey has, in its 2024 report “A New Future of Work: The Race to Deploy AI and Raise Skills in Europe and Beyond”, a term for what is happening to Romania: “double jeopardy.” It applies to countries that are simultaneously the most exposed to automation and the slowest in AI adoption. The mechanism is simple and brutal: they lose jobs without gaining productivity. They pay the cost of the transition without receiving its benefits. They lose on both fronts at the same time.
The McKinsey report further estimates that the transition across Europe is not uniformly distributed. Countries with weak reskilling systems and fragile digital institutions will absorb the shock without capturing the benefits, the authors explicitly warn. This forecast is, moreover, grounded in valid historical precedents.
Romania ticks every box for this trap. Its economic structure is concentrated in categories with a high risk of automation. It has one of the lowest levels of AI adoption in the EU. It has an underfunded and unreformed vocational training system. And it has no institution to actively manage the transition. The double trap is not a theoretical risk for Romania. It is simply a description of the current situation.
Mustafa Suleyman names this dynamic in The Coming Wave (2023) “the containment problem.” In short, institutions advance far more slowly than technology. Romania illustrates this asymmetry perfectly: technology enters the economy through the private sector via a handful of early adopters and local champions, while the institutional infrastructure that should manage the transition remains stuck in a 20th-century logic and in dusty archives full of paper-clipped folders.
What AI Automates First and Which Jobs Are at Greatest Risk
The World Economic Forum, in its Future of Jobs Report 2025, the most comprehensive annual study on the global labour market, based on data from over 1,000 employers representing 14 million employees across 55 economies, lists the occupational categories with the highest risk of automation in the 2025–2030 horizon:
- Data entry operators and document processors
- Accountants, auditors, and mid-level financial analysts
- Administrative clerks in the public and private sectors
- Call centre operators and customer service staff
- Financial and banking transaction processors
- Entry-level and mid-level data analysts
- Secretaries and administrative assistants
These are not marginal categories in the Romanian economy. They are part of the backbone of the urban middle class built over the past 20 years: the BPO and shared services sector in Cluj, Iași, Bucharest, and Timișoara; banking and multinational back-office operations; central and local public administration; accounting and audit firms serving Romania’s tens of thousands of SMEs.
We have no official calculation of the exposure. No one has produced one, and this is itself a governance problem, not a technical one. INS data on occupational structure, when placed alongside the WEF taxonomy and the IMF’s classification of AI exposure by job category, suggests that between 1.2 and 1.8 million Romanian employees belong to categories with a high risk of automation in the 2025–2030 horizon. As a proportion of the total workforce, this is one of the highest levels of exposure in Central and Eastern Europe.
One further point worth emphasising: being in the risk zone is different from automatically losing one’s job tomorrow. But the nature of that work will be radically and rapidly transformed, the economic value of that work will compress, and without proactive reskilling and parallel AI adoption, these employees will become increasingly less competitive relative to their counterparts in countries that are actively tackling the transition. Doug O’Laughlin describes this bluntly: “$10–$25k replaces the job of a $120,000-a-year analyst. That doesn’t mean the analyst disappears tomorrow, but their value is in immediate collapse.”
The 2024 IMF report, “Gen-AI: Artificial Intelligence and the Future of Work,” produced by eight of the Fund’s economists, refines this analysis and introduces an essential distinction between “AI-complementary” economies, those with many highly skilled knowledge workers, where AI augments productivity, and “AI-exposed” economies, those with a high proportion of routine cognitive work, where AI directly substitutes for labour. Romania firmly falls in the second category. Not only because of the educational level of the workforce, one of the most commonly cited arguments, but more so because of its economic structure: we have built a mid-tier services economy, precisely the segment that AI automates fastest and most completely.
The Adoption Paradox: Why the Most Exposed Adopt the Least
Eurostat shows that in 2025, 1 in 6 Europeans already uses generative AI at work, a significant increase from the previous year. At the opposite end of the European ranking sits Romania, with approximately 5% professional use of generative AI, according to the same report. This figure is linked to public policy, or, more precisely, to the absence of it.
.png#file){kind=link}
There is a perverse logic to this paradox. The countries most exposed to automation are also those that adopt AI most slowly precisely because they lack the governance structure to force and facilitate the transition. Without public policies that stimulate adoption, without reskilling programmes, without clear signals from the state that this technology is a national priority, people stay put. Companies delay. SMEs lack the resources to experiment on their own. And the public administration, which could be the first large-scale adoption engine and could signal the direction for the entire economy, remains trapped in a paper-folder logic.
McKinsey warns that the difference in outcomes between European countries is a consequence of public policy choices. “Choices made today can reignite productivity growth,” the authors write, and the difference between the active and the passive scenario is a spread of 3% annual growth versus 0.3% — ten times less. France, Estonia, and Denmark have made those choices. Romania still hesitates, even as its economic performance in 2024–2026 looks more like stagnation than growth.
Research published in April 2026 by Eduardo Amaral, Mijail Naranjo-Zolotov, and Fernando Bação, the first peer-reviewed study measuring AI readiness in the e-governance of all EU member states, using harmonised Eurostat and European Commission eGovernment Benchmark data, confirms and calibrates this picture. The authors operationalise the degree of readiness for AI adoption along two dimensions: institutional factors (the stability of administrative arrangements, decision-making transparency, the level of trust in government) and organisational factors (the capacity to deliver integrated digital services, interdepartmental coordination, adaptability to new technologies).
“Institutional and digital disparities between member states create an uneven playing field from the outset,” the study concludes. “Institutional and organisational factors measurably contribute to differences in AI readiness between EU member states.” Romania is in the maximum-risk cluster on both dimensions. The chart is sobering. That is precisely why we include it below.
Engels' Pause, Romanian version
You have probably heard of the Industrial Revolution as a clear success story. But between 1780 and 1840, productivity exploded while workers’ wages stagnated or even declined. The gains went almost exclusively to capital owners. Workers paid for the transition with decades of stagnation. The benefits came much later, and not automatically, but as the result of building new institutions: trade unions, social legislation, universal public education. This episode in economic history is known as Engels’ Pause.
Doug O’Laughlin argues, in the same excellent article, that AI will produce a new Engels’ Pause, but with a crucial difference in temporal scale and social impact: “The salary of a professional will collapse to the level of an unskilled worker. And new graduates? They have no chance.” The speed is incomparably faster than in the Industrial Revolution. Great Britain had 50–60 years to build the institutions to manage the transition. We may have perhaps 5, he says. For an even sharper angle, the Substack post “Riding the Bomb: AI’s War on White-Collar Jobs” draws a similar parallel between Engels’ Pause and the declaration by Fiverr’s CEO: “It doesn’t matter whether you are a programmer, designer, product manager, data scientist, lawyer, customer care professional, salesperson, or financial analyst, AI is coming for you.”
There is a second difference, more important and more specific to the Romanian context. Great Britain in the 1800s was the centre of the global empire: rich in resources, in concentrated human capital, and with a political class that, even if it reacted slowly, did eventually react. It built the Corn Laws, the Factory Acts, the Education Act. It created the institutional mechanisms that transformed a painful pause into a transition with a positive outcome.
Romania is entering the equivalent of this period without any institutional safety net. Without an institution responsible for AI. Without a reskilling strategy for exposed categories. Without a consistent programme for AI adoption in the public sector to signal a direction. Without a single budget dedicated to managing the labour market transition in the context of AI. The pause, under these conditions, can be more than a temporary stage. It can become a trap from which escape grows increasingly difficult with every month of hesitation.
There are also voices that use a different tone in this chorus of pessimists. Emily Bender and Alex Hanna, in The AI Con (2024), warn that part of the catastrophic narrative about “job disappearance” serves the interests of big tech companies, creates public panic that justifies hasty political decisions and massive budgets directed towards private solutions. The message is not that AI’s impact on the labour market is non-existent, but that it must be measured empirically, not absorbed from the press releases of major companies or from the maximal projections of global consultants.
This nuance is relevant for Romania: we do not need panic, we need data. Unfortunately, the data we already have, 70.65% exposure to automation, 3.07% AI adoption in companies, 5% of employees using AI versus nearly 15% the EU average, are sufficient to justify urgency without resorting to apocalyptic scenarios.
Institutional Discontinuity as an Involuntary Strategy
Professor Bernd Wirtz of the German University of Administrative Sciences Speyer, one of Europe’s most prestigious public administration research institutions, identifies in Digital Government: Strategy, Government Models and Technology (Springer, 2022) a structural trap specific to states with fragile governance, which he calls “discontinuities due to short-term legislature orientation”: the interruption of any strategic continuity at every electoral cycle, caused by the short-term orientation of political actors.
The mechanism is well known in Romania, even if we have no academic term for it. And it is extremely relevant during this period of heightened political instability. Every new government brings a new digital vision. Every minister of digitalisation or the economy starts with a blank slate. As of 1 May 2025, the leader of this ministry was Bogdan Ivan (subsequently moved to Energy), then Radu Miruță (moved to Defence but also Transport, to cover an interim role), and now Irineu Darău, who, following the recently passed motion of no confidence, will not remain in post for more than 45 days. Today we do not know who will be minister of MEDAT on 1 July, where they will be moved next, or how long they will hold the role.
As a result, ongoing projects are in many cases abandoned and teams reshuffled. The National Strategy on Artificial Intelligence (SN-IA) 2024–2027 was approved in July 2024, a decent document, aligned to European standards, but without an explicitly allocated budget and without an institutional structure with a clear implementation mandate. It is, in essence, a strategy for which no one is explicitly responsible.
Wirtz argues that this discontinuity is not accidental and cannot be resolved through episodic political will. It is a structural problem that requires structural solutions: institutions with a multi-governmental mandate, funding protected from the annual budget cycle, and accountability mechanisms that function independently of political rotation. “Digital government cannot be successfully implemented without long-term strategic commitment that transcends individual government mandates,” the author stresses.
Before we consider this an exclusively Romanian problem, it is worth looking at the contrast. Estonia built X-Road, the infrastructure that makes digital government possible, across three governments and two decades. It did not stop with every political rotation. Denmark built its AI capacity in public administration over a period of seven years, with institutional continuity regardless of the government’s political colour. These models work because their politicians understood and solved the discontinuity problem through institutional architecture.
Romania has not yet done so. And, I would argue, today’s leaders understand the issue in the terms described above and do not view digitalisation, as with other topics, as a multi-governmental chapter with budgets and teams planned over the medium and long term. We discussed this need at length in the context of the report “The Digital Governance Framework in Romania,” produced together with the Estonian experts from Digital Nation last autumn. In the absence of this kind of approach, any national strategy remains a well-intentioned document with no realistic chance of implementation.
The Closing Window: What Recent European Research Shows
The Amaral et al. (2026) study delivers a conclusion that any political actor should read carefully: the gap between states in AI readiness is not built at the moment of AI adoption, but beforehand, during the period when institutional and digital capacity is, or, in our case, is not, being constructed.
In plain terms, there is no longer a future stage at which Romania can “jump” to the top of the ranking through the late adoption of better tools. This is perhaps the point I would stress most. We have all used the example of cable and broadband internet penetration in Romania two or three decades ago, which brought us from the bottom of European rankings to among the leaders. It was an entirely new technology and infrastructure. We had no “hegacy” to protect, goes the argument, one still used today.
Only in the case of AI, this argument does not hold. The necessary institutional capabilities, integrated digital services, interdepartmental coordination, the ability to adapt to new technologies, interoperable structured databases. All of these take years to build, at minimum. And every year of inaction is not a neutral year. It is a year in which the distance from the states that are acting grows exponentially, because data volumes and cross-border exchanges grow proportionally for others while growing slowly for us.
The study also identifies an important asymmetry: states with high AI readiness in e-governance generate a spill-over effect into the private economy. When the state adopts AI visibly and at scale, it signals to the business community that the technology is mature and that adoption is strategically important. It stimulates by example, creates demand for local AI talent, and develops the ecosystem. States with low AI readiness in e-governance, conversely, transmit a negative signal — that the technology is not a priority, that risks outweigh benefits, that there is no clear framework.
Romania is actively transmitting the second signal. It transmits it primarily through absence, even as public declarations are full of buzzwords.
Amaral et al. conclude that “the window of opportunity for closing AI readiness gaps in the EU is temporally limited”. Countries that do not act within the 2025–2027 horizon will enter the second wave of global AI adoption from a structurally inferior position, with recovery costs significantly higher than the costs of preventive action now. For a country in a period where economic growth is closer to 0% and the deficit is large, this verdict is stinging.
The Leadership That Is Missing and Why It Matters More Than Strategy
Professor Miriam Lips of Victoria University of Wellington, a recognised academic voice on digital governance, states directly in Digital Government: Managing Public Sector Reform in the Digital Era (Routledge, 2020) a conclusion that should be displayed at the entrance to every ministry in Romania: “Leadership is one of the most critical domains for success in digital governance. Public managers will not lose their jobs because of AI, but their roles will involve managing highly complex, interconnected, and hierarchically structured governmental digital systems”.
In other words: managers in the central administration have today neither the mandate, nor the capacity, nor the tools to manage the complexity that AI introduces into the systems they oversee. First and foremost because no one has defined that this is their task. There are no dedicated teams and support structures, there is no national-level leader responsible for this issue who is accountable for solving it.
Lips argues that digital transformation in the public sector fails, when it does, almost invariably not because of the technology. The technology exists, is available, and is often even financed through European funds, but because of “the absence of sustained leadership that integrates the vertical, horizontal, and socio-technical dimensions of change.” Vertical means coordination between levels of government. Horizontal means inter-ministerial coordination. Socio-technical means integrating the real needs of citizens and civil servants into the design of solutions. Without a leader with an explicit mandate across all three, each dimension is addressed piecemeal, incompatibly, and inefficiently.
Romania has demonstrated this theorem multiple times over the past 15 years of failed digitalisation. Information systems that do not talk to each other. Parallel databases for the same data sets. PNRR projects built on architectures incompatible with existing systems. Not from ill will, we repeat this here too; but from the absence of a leader with integrated authority and responsibility. A strikingly similar conclusion can be found in our report produced together with Digital Nation, as well as in various OECD reports on digital governance in Romania.
AI amplifies this problem exponentially, because the speed and scale of the transformations required completely exceed the capacity for ad-hoc coordination. Someone needs to be accountable. And in Romania, on this decisive issue, no one is.
Karen Hao documents in Empire of AI (2025) that the absence of governance creates a vacuum actively occupied by the interests of the companies developing the technology. When the state has no strategic architecture and the national AI agenda is written by no one, it ends up being implicitly written by platform vendors. Romania, with 3.07% of firms using AI and without an institution responsible for strategic direction, is not outside the game. It is in the game, but playing a hand dealt by others.
A World Economic Forum report from April 2026, a study assessing governmental readiness for adopting agentic AI, adds another warning: Gartner estimates that over 40% of agentic AI projects will be cancelled by 2027 due to insufficient strategic planning. In this case, others’ problem could be an opportunity for us. Romania has not yet begun this chapter, but if it starts without a strategic architecture and without an institution responsible for prioritisation, it will produce, with high probability, exactly the kind of failure we know well: fragmented projects that drain resources and erode institutional trust in technology for a decade.
France offers, right now, proof that this problem is not unique to Romania. In mid-April 2026, the National Agency for Secure Documents, the institution that manages French passports, identity cards, and driving licences, suffered a breach that exposed the data of 11.7 million citizens. Prime Minister Lecornu immediately announced an emergency package: 200 million euros, 5% of each ministry’s digital budget allocated to cybersecurity, a new state digital authority. The response was swift, the resources were there, and the political will was visible. France has ANSSI, one of Europe’s most powerful cybersecurity agencies, a national strategy published in January 2026, and a Campus Cyber with 1,800 experts.
And yet, if France, with all this apparatus, discovers that its critical systems are vulnerable, that responsibility is unclear between ministries, and that, in Lecornu’s words, “it is not ANSSI’s job to ensure that ministries are properly protected”. What can we say in Romania’s case? In Bucharest today, basic digitalisation, cyber defence capacity, and the AI governance conversation are all being built simultaneously from near-scratch, starting from an institutional void that does not permit the luxury of inaction even for a single quarter. France’s problem is one of execution. Ours is one of architecture, the kind that needs to be built first, so that execution can take place at all.
Niklas Lidströmer adds in The AI Ideal (2026) a complementary critical perspective: “Aidealism”, the belief that AI will automatically solve societal problems if left to operate freely, has colonised public policy discourse in many European countries. Romania does not suffer from aidealism. It suffers from the absence of any coherent ideology about AI. Which is, paradoxically, even more dangerous: you cannot correct a wrong vision if no vision exists. We wrote about this context and the absence of vision, in different words, with different quotes and different references, a few weeks ago. We will do so again. The stakes are that high.
What Edge Institute Proposes
Perhaps our conclusion is predictable. We make no apology for that. Or at least I don’t. Sometimes simple things are simple because they are true.
You cannot manage a risk you do not measure. You cannot measure a risk if you have no institution responsible for collecting data and drawing conclusions. You cannot build a technological transition strategy without institutional continuity beyond the electoral cycle. And you cannot do any of this without a leader with a clear mandate, authority, and accountability.
Romania does not know how many employees are in the risk zone. It has no reskilling programme, no map of vulnerable jobs, no AI adoption policy for SMEs. And it has not a single institution with the mandate to produce any of these. Zero.
Keach Hagey’s The Optimist (2025), the biography of Sam Altman and OpenAI, offers an indirect warning for Romanian decision-makers: the speed at which AI companies build relationships with governments that act quickly (the US, UK, Japan, UAE) means that states that delay defining a national position do not receive a seat at the negotiating table, they receive a finished product that they can adopt or reject, with no capacity to influence it. The national AI consensus we are proposing is not a strategic option among others. It is a chance to participate in writing the rules, or at least in adapting them locally. Otherwise, the default option will remain reading them after they have been written by others.
The IMF, McKinsey, WEF, and researchers Amaral et al., Lips, and Wirtz have produced the analytical tools. There are many more; we chose only a few for this article. The methodology exists. Best-practice examples exist in Europe. Estonia, Denmark, Finland, and France have demonstrated that the transition can be actively managed. What Romania lacks is the political will to put the issue on the table and to appoint someone responsible for answering it.
Edge Institute’s proposal is the same as in the previous article in this series, or as in the broader context of digital transformation, but with an additional urgency added by every passing month:
An AI Adviser at the Presidential Administration, where the natural mandate extends beyond the governmental cycle and where longer-term visions can be assumed, with an explicit role of coordinating national policies on occupational transition and AI adoption, paired with a GCIO office in the team of the Deputy Prime Minister dedicated to digitalisation, with a concrete mandate, allocated budget, and public accountability against measurable targets.
Not in 2028, in the next governance programme. Not in 2027, when it appears on the radar of yet another new minister. Now. Because the window that Amaral et al. describe is closing rapidly, McKinsey’s double trap deepens with every quarter of inaction, and Engels’ Pause in its Romanian version has no institutional mechanism to transform it into a decent transition. Such a dedicated authority for AI governance was envisaged in the National AI Strategy, but has not yet been operationalised.
You cannot lose a race you did not enter. You can try again next time, if the race is still open. In the context of AI, Romania’s leaders may come to understand that a prolonged absence from a technological competition produces not just another place in the back of the pack in a new ranking, it produces a restructuring of the economy around you, without you, by the rules of others.
Today, 6 May, the Presidential Administration is hosting a roundtable at Cotroceni entitled “The New Architectures of Digital Democracy”. It is a sign that the conversation is beginning to exist at the level at which it must exist. But a conversation, however well-intentioned, does not replace a decision. Romania needs, alongside a new space for dialogue, the first institutional framework in which the conclusions of these dialogues become policies with an owner, a budget, and a deadline. This article is published on the very day of that event, not by coincidence, but in the hope that those attending will leave with more than a ticked agenda item.
The absence of governance is a choice. Don’t you want to change it?
Other countries are losing the race. Romania has not truly shown up at the starting line. There is no outcome harder to correct than absence, because absence leaves no traces, generates no lessons, produces no political pressure. It passes unnoticed until it is too late to notice anything useful.
Through inaction, day by day, ministry after ministry, strategy after strategy without an owner, someone is closing this window of opportunity. Absence, today, is a choice. And like any choice, it can be changed, but only if there is someone responsible for changing it.