The AI conversation in CEE is beginning to shift from technology to systems. This is what sat at the center of conversations during the Regional AI Valley conference in Bucharest, where speakers from technology companies, financial institutions, and public agencies from Romania and Moldova discussed how AI is beginning to intersect with sectors that operate on very different timelines: private capital, government institutions, and critical infrastructure.
TL;DR – it seems we’re entering a more complex phase of AI adoption, one in which infrastructure, governance frameworks, and institutional capacity may prove more decisive than the technology itself. In practice, this means that the next stage of AI development in CEE will likely depend less on access to models and more on whether countries can build the surrounding systems required to deploy them responsibly and at scale: interoperable digital platforms, compute infrastructure, regulatory clarity, and teams capable of operating increasingly complex AI systems.
Private capital may move faster than public institutions
One of the recurring tensions discussed throughout the event concerned the speed mismatch between private investment and public sector readiness.
According to Victoria Zinchuk, EBRD Head for Romania, Romania already possesses some of the structural advantages that attract AI-related investment:
“The private sector will come to Romania regardless of the readiness of the public sector,” she said, pointing to the growing pipeline of potential projects linked to energy and digital infrastructure.
This idea dominated the conference, with talks of a broader pattern across the region, where AI strategies often exist on paper while implementation capacity continues to evolve.
Digital government infrastructure as an enabling layer
If infrastructure readiness remains uneven across the region, Moldova offered an example of what long-term institutional investment in digital government can produce.
Nicoleta Colomeet, Director at Moldova’s e-Governance Agency, described how the country’s digital transformation strategy, first adopted in 2006, gradually evolved into an interoperable digital ecosystem.
“Back in 2006 Moldova decided this is a concept for digital transformation,” she explained, referring to the decision to develop a modular digital infrastructure capable of supporting future services.
The result today is what she described as a “platform of platforms” model connecting public institutions and private services.
“We are incredibly interoperable,” Colomeet said, noting that Moldova’s infrastructure now facilitates over three million data exchanges per day.
Such systems increasingly represent the underlying layer on which AI-powered public services can realistically be deployed.
The overlooked constraint: AI infrastructure
Beyond policy frameworks, speakers repeatedly returned to a more practical question: whether organizations are technically prepared to run AI workloads at scale.
For Vlad Biristeica, Cloud & AI Lead at Cisco Romania & Moldova, the conversation around AI readiness often misses the point.
“As a company, your objective is not to build infrastructure,” he said. “Your objective is to run AI services and help the citizens or the business.”
Yet achieving that objective depends on the availability of infrastructure capable of supporting intensive compute workloads and large-scale data operations.
“AI is everywhere,” he said about a phrase often used internally at Cisco: “But is your data center AI-ready?”,
suggesting that the technical foundations of AI adoption are still heavily uneven in the region, even as interest in the technology continues to accelerate.
Finance: clear business cases, uneven adoption
The financial sector provided a more operational perspective on AI deployment. Applications such as fraud detection, risk modeling, and customer analytics already offer measurable value for banks and payment providers. But adoption remains uneven across institutions.
For Alexandru Mihalache of Visa, the key driver remains straightforward.
“Put simply, it’s the business case,” he said.
When the cost of implementing an AI model is clearly outweighed by the potential reduction in fraud or operational losses, adoption becomes easier to justify.
Yet economic logic alone does not guarantee implementation.
According to Gevorg Safaryan of EarlyOne, an Armenian-based customer flow management system, organizational capacity remains one of the biggest constraints.
“The biggest issue is the lack of capacity within the teams,” he said, describing how many financial institutions still lack internal expertise to deploy and evaluate advanced data systems. We have 40 banks as clients… but only about 10 are really using the data in the right way to evaluate different situations.”
The panel concluded with AI adoption depending as much on internal skills development as on technological availability. Which brings us to our next topic:
Talent remains the central variable
If infrastructure and governance form the structural layer of AI adoption, talent remains the operational one.
During a session focused on practical implementation, several speakers returned to the same point: organizations must rethink how they build teams in an AI-driven environment.
For Georgeana Trofim, VP Engineering & Data at Booking, the most important investment remains unchanged.
“The right people,” she said. “The forward-looking dreamers.”
Others emphasized the need for teams capable of translating AI capabilities into business value.
“Finding the right people who are capable of understanding AI so we can double down on the way we create and capture value,” said Andrew Taylor, Managing Director at Connect CEE.
Some participants also highlighted the speed at which AI tooling itself is evolving.
“I would invest in Claude skills as quickly as possible,” said Paul Chirita, Director of Machine Learning Engineering at Adobe, referencing the rapid emergence of AI-native development workflows.
Entrepreneur Andrei Pitis, CEO of Genezio, suggested that founders starting today may approach team building differently altogether.
“If I were starting today without the network I’ve built, I wouldn’t try to do everything myself,” he noted.
Instead, he would rely more heavily on AI assistants while focusing on collaborators who bring complementary expertise.
AI systems require quality human judgment
Despite the rapid expansion of AI applications, several speakers emphasized that automation does not eliminate the need for human oversight. On the Finance track, Aliona Stratan from maib stressed that organizations must retain the analytical capacity to understand and challenge the outputs generated by AI systems.
“When developing an AI model, if something goes wrong we should be able to go back and understand what should be changed, improved and monitored. If we don’t bring people who understand what is behind the result of a model, this could become dangerous.”
As AI systems move deeper into decision-making processes, the ability to interrogate their outputs becomes as important as the models themselves. Automation may accelerate analysis, but it’s far from replacing human judgment.
One term often talked about in the context of complex AI models that take over decision-making processes is moral deskilling, describing the slow erosion of people’s ability to question automated outcomes. Preventing that will require keeping teams firmly responsible for supervising AI systems, which in turn means investing across organizations in transferable skills and meta-competences that allow teams to interrogate, interpret, and challenge the systems they deploy.
From experimentation to infrastructure
Taken together, the conversations at Regional AI Valley suggest that Central and Eastern Europe is entering a new phase of AI adoption. The challenge we’re facing now lies in building the surrounding ecosystem: interoperable digital infrastructure, technical capacity within organizations, regulatory clarity, and a workforce capable of operating increasingly complex AI systems.
In that sense, the next stage of AI in CEE will likely be less visible than the first. It won’t be with the shiny impressive technical models we’ve grown to love, but by the much less glam work of actually building the institutions, systems, and expertise required to use them sustainably.
