The JRC study examined 29 initiatives across the five countries, combining a review of policy and academic literature with interviews at policy, institutional and classroom level. Its central finding: AI uptake in VET is condition-driven. Progress depends less on which applications are available, and more on whether an enabling environment exists. As an exploratory qualitative study of selected initiatives, it illustrates key patterns without claiming statistical representativeness. 

Four elements repeatedly proved decisive: solid governance capacity, systematic teacher competence development, trusted digital infrastructure, and stable support from intermediary networks linking VET institutions with industry. Where these are in place, AI-enabled VET can scale. Where they are missing, adoption stalls at the pilot stage, dependent on individual "pioneer" teachers rather than system-wide practice. This makes AI adoption a question of system desing and institutional capacity, rather than a race to deploy even more tools.

Rather than counting how many AI tools are in use, the study analyses adoption as a process spanning three levels: macro (policy and regulation), meso (intermediary networks and coordination bodies) and micro (training centres, teachers, learners and companies). Macro-level strategies only translate into practice when meso-level intermediaries — such as sector networks or centre-industry agreements — provide coordination, training and the time needed for teachers to experiment. 

This layered approach helps explain why implementation remains uneven despite widespread policy ambition. All five countries frame AI as an essential, human-centric complement to educators — not a replacement — and link their strategies explicitly to the EU AI Act, which classifies some educational AI applications as "high-risk" and imposes obligations on transparency, data quality and human oversight. 

Where conditions allow, AI is already supporting more personalised tutoring, helping bridge language and learning barriers, and easing administrative workload for teachers. The study also observes a shift toward process-oriented assessment - oral tests, reflective portfolios, continuous feedback - designed to safeguard academic integrity as AI tools become more capable. 

Persistent barriers remain, however: vendor lock-in, unpredictable token-based pricing for AI services, uneven AI literacy among staff, and a lack of clear legal and ethical guidance for institutions handling learners' data. The report argues that closing these gaps requires formal teacher accreditation in AI competences, protected time for educators to innovate, and sustainable funding that goes beyond short-term project cycles. Taken together, these prerequisites turn AI integration into a long-term capacity-building effort rather than a series of isolated pilot projects. 

Sustainable AI integration in VET requires a shift from technology-centric investment to human-centred outcomes: building the workforce's capacity to use AI critically and safely, rather than simply expanding the number of tools in use. In this sense, AI becomes part of a broader digital transformation of VET systems, where governance, compliance and teachers' development are as crucial as the technologies themselves.