Personalized learning (PL) aims to provide an alternative to the one-size-fits-all approach in education by tailoring instruction, pace, methods, and content to individual learners' needs. AI-driven PL solutions have shown effectiveness in enhancing learning performance, but their alignment with modern educational goals remains inconsistent. This paper examines the characteristics of AI-driven PL solutions in light of the OECD Learning Compass 2030 goals, identifying gaps between modern educational objectives and current PL approaches. While PL technologies support learning processes, the vision of educational experts extends beyond technological tools, requiring a holistic change in the educational system. The paper explores the potential of large language models, such as ChatGPT, and proposes a hybrid model that blends AI with a collaborative, teacher-facilitated approach to personalized learning. The OECD Learning Compass 2030 emphasizes three main goals of education: focus on general competencies, developing learners' agency, and building on the Anticipation-Action-Reflection (AAR) cycle. These goals highlight the need for deeper cognitive engagement, learner agency, and the development of skills such as critical thinking, collaboration, and complex problem-solving. While AI-driven PL systems have shown benefits in improving learning efficiency and outcomes, they often focus on domain-specific knowledge and may neglect broader competencies. Additionally, PL systems may limit learner agency and self-regulation, which are essential for lifelong learning. The paper also discusses the challenges of PL systems, including individualization, focus on performance, domain-specific knowledge, limited agency and learning skills, and engagement and motivation from gamification. While gamification can increase engagement, some elements may decrease intrinsic motivation. The paper argues for a hybrid model that combines AI with teacher-facilitated learning to support both learners and teachers in personalizing learning content in scientifically proven ways. This model emphasizes the importance of self-regulation, collaboration, and cognitive activation of learners' minds. The future of formal education is likely to be a hybrid model of reciprocal interactions between humans and AI, rather than a fully automated system. This approach requires a system-wide effort to support self-regulated learning, with resources directed towards training children in SRL skills as early as possible. The paper concludes that AI-driven PL should be guided by educational goals and appropriate technology, rather than the other way around.Personalized learning (PL) aims to provide an alternative to the one-size-fits-all approach in education by tailoring instruction, pace, methods, and content to individual learners' needs. AI-driven PL solutions have shown effectiveness in enhancing learning performance, but their alignment with modern educational goals remains inconsistent. This paper examines the characteristics of AI-driven PL solutions in light of the OECD Learning Compass 2030 goals, identifying gaps between modern educational objectives and current PL approaches. While PL technologies support learning processes, the vision of educational experts extends beyond technological tools, requiring a holistic change in the educational system. The paper explores the potential of large language models, such as ChatGPT, and proposes a hybrid model that blends AI with a collaborative, teacher-facilitated approach to personalized learning. The OECD Learning Compass 2030 emphasizes three main goals of education: focus on general competencies, developing learners' agency, and building on the Anticipation-Action-Reflection (AAR) cycle. These goals highlight the need for deeper cognitive engagement, learner agency, and the development of skills such as critical thinking, collaboration, and complex problem-solving. While AI-driven PL systems have shown benefits in improving learning efficiency and outcomes, they often focus on domain-specific knowledge and may neglect broader competencies. Additionally, PL systems may limit learner agency and self-regulation, which are essential for lifelong learning. The paper also discusses the challenges of PL systems, including individualization, focus on performance, domain-specific knowledge, limited agency and learning skills, and engagement and motivation from gamification. While gamification can increase engagement, some elements may decrease intrinsic motivation. The paper argues for a hybrid model that combines AI with teacher-facilitated learning to support both learners and teachers in personalizing learning content in scientifically proven ways. This model emphasizes the importance of self-regulation, collaboration, and cognitive activation of learners' minds. The future of formal education is likely to be a hybrid model of reciprocal interactions between humans and AI, rather than a fully automated system. This approach requires a system-wide effort to support self-regulated learning, with resources directed towards training children in SRL skills as early as possible. The paper concludes that AI-driven PL should be guided by educational goals and appropriate technology, rather than the other way around.