What to Do: A Framework for a Managed AI Transition in Kazakhstan’s Schools

We have come a long way. We began with a simple but uncomfortable thesis: the question is not whether to introduce artificial intelligence into school, but how to manage a process that is already under way. We then looked at global trends and became convinced that this will not simply blow over. We heard the voices of our own education system — hundreds of teachers, parents, principals, methodologists, and students. We examined other countries’ mistakes and saw that no country has yet solved the issue. We made an honest diagnosis of school itself and saw the risk that AI may amplify its problems rather than solve them. Finally, we reached the hardest knot: the data and platforms on which everything rests.

All this led to one question, the same question we posed in the first material: not “can AI be implemented,” but “how and why.” It is time to answer it. This final material is about what a managed, staged, and human-centered transition framework might look like. Not a set of slogans, but an architecture for decision-making.

If everything we have learned is compressed into its densest core, five supporting principles emerge. This is not a declaration of intentions, but the managerial nucleus on which the entire policy should stand.

First: the state governs the balance, not the technologies. It is not the state’s role to develop every algorithm and control every classroom. Its role is to maintain equilibrium among three forces that otherwise develop out of sync: fast-moving technologies, lagging institutions, and vulnerable people with their values. Balance, not micromanagement.

Second: the teacher is the central subject of the transition, not its victim. We have seen in South Korea what implementation over the heads of teachers leads to. The teacher must be a co-author of change, not the person onto whom it is imposed by order. If the teacher sees AI as an ally, the transition can work. If it is perceived as an obligation, even a well-funded project will fail.

Third: human development matters more than digital efficiency. When there is a choice between “faster and cheaper” and “more humane and meaningful,” the priority belongs to the latter. School exists not to optimize indicators, but to grow a human being.

Fourth: experiments are separated from mandatory standards. Schools and teachers need space to try, make mistakes, and learn — but within common frameworks of safety, ethics, and data protection. Freedom of experimentation and shared standards do not contradict each other when they are carefully distinguished.

Fifth: national identity is embedded in digital infrastructure. The Kazakh language, culture, and context should not be an add-on placed over foreign solutions, but part of the foundation itself. Otherwise, global algorithms will quietly rewrite our school according to someone else’s model.

These five principles are not foresight and not a list of initiatives. They are already an architecture of state policy, compressed to its essence.

A roadmap grows out of these principles. The first thing to understand about it is that it is not a sprint, but a march in three stages, each with its own main task. Trying to jump over stages is exactly the mistake on which others have burned themselves.

Stage one, 2025–2026: put things in order, not scale. This is critical. The key task of the next two years is not to push AI into every school, but to organize the chaos and create a base. This includes launching protected pilots for preparing materials, checking work, and management analytics; forming requirements for EdTech providers and independent audit mechanisms for their solutions; ensuring basic infrastructure — internet and devices; beginning mass but practice-oriented retraining of teachers; developing ethical principles and school codes for AI use; and experimenting with hybrid assessment models in which AI checking is complemented by oral defense. One point should be highlighted: at this stage, forced scaling and rigid KPIs are unacceptable, because they produce formalism, burnout, and resistance.

Stage two, 2027–2030: assemble and standardize. When the base is ready, the system can move from scattered experiments to systemic solutions: creating a national educational AI platform with digital profiles and adaptive trajectories; developing a national language model that accounts for the Kazakh language and values; introducing personalized trajectories and digital portfolios; institutionalizing new roles such as AI curator and educational data analyst; and developing professional communities and horizontal exchange of practices. This is the stage at which the system matures.

Stage three, after 2030: from technologies to the human being. On this horizon, the focus shifts from tools to the essence of education: the final move away from the exam as the dominant form of assessment; assessment of motivation, resilience, and the ability to cooperate; creation of “AI-free zones” for developing deep, long cognitive skills; and a new role for the Ministry — governing an entire educational AI ecosystem and maintaining the balance between digital and live experience. This is the horizon of maturity, and it cannot be reached by skipping the first two stages.

Running through this entire framework is a rethinking of the role of the state. This should be stated separately, because confusion arises here most often.

The state in the AI transition is not a foreman laying every brick. It neither must nor can develop all solutions, operate every tool, or supervise every classroom. Trying to carry everything on itself is doomed: the system is too large and the technologies too fast.

But the state has a role that no one else can play. It is the architect of the frameworks within which all others act. The guarantor of values that cannot be sacrificed to efficiency. The moderator of pace, preventing the process from overheating or freezing. And the protector of the vulnerable — first of all children and teachers — those who risk the most in this transformation. The state does not govern technologies. It governs the balance and holds what cannot be left to the market or chance.

Principles and stages explain “what.” Now to the “how”: what concrete mechanisms can hold this construction together. AI transformation is moving along several uncoordinated tracks at once — market, state, project-based, and spontaneous — and none of them alone ensures governability. Linkages are needed. There are three of them.

The first is an association of EdTech market players as a mechanism of self-regulation. Kazakhstan’s educational AI market already exists, but remains fragmented: there are no common professional benchmarks, agreed approaches to data, or shared understanding of pedagogical responsibility. An association is not a lobbying structure, but a tool of self-regulation: developing minimum professional standards, aligning positions on sensitive issues such as data, assessment, and the role of the teacher, creating a permanent channel of dialogue with the state, and increasing trust among schools and parents. It does not replace state regulation; it removes part of the burden by moving conflict-prone questions into the professional field.

The second is three integrative projects that function as infrastructure rather than grants. Today there is no space between the market, the state, and the school where solutions can be tested safely and data on real effects accumulated. Three instruments can create it. A transformation accelerator — a program to support those who implement AI locally: principals, management teams, and regional coordinators; it teaches them to work with risks, teams, and data, reducing the likelihood of chaotic decisions. Learning circles — a format of horizontal exchange among teachers and schools, a protected space where practices, including unsuccessful ones, can be discussed and scattered experience turned into shared knowledge without being pushed to premature scaling. Open dashboards — a tool for transparent monitoring that tracks not individual schools for punishment, but systemic effects: inequality, teacher workload, changes in results, and social risks, so that policy can be corrected early, while mistakes are still cheap.

The third is a parliamentary contour. Given the social sensitivity of the topic and the scale of possible consequences, it makes sense to create a special mechanism on the parliamentary platform for coordinating the AI transformation of school education. Its role is not operational management, but monitoring key risks, providing public and political legitimation of decisions, interdepartmental coordination, and regular public discussion. Such a contour distributes responsibility among executive power, the market, and the professional community without concentrating it in one point and without increasing administrative pressure.

Projects deserve separate attention — the many initiatives proposed by participants in our studies, from teacher training and personalized learning platforms to cyber hygiene courses from the first grade, AI leadership schools for principals, palaces of schoolchildren as laboratories of the future, and the idea of making Kazakhstan a global center for AI ethics in education.

It is important here not to make a methodological mistake. Projects are not ready-made answers and not “best practices” that should be rolled out across the country as quickly as possible. They are transitional mechanisms, a way of working with uncertainty. Their function is to test hypotheses, identify sensitive zones, and translate attractive visionary ideas into a manageable format. Expecting quick scalable effects from them means misunderstanding their nature.

The practical conclusion follows from this: the project layer should be preserved as a space for experimentation and navigation, not as a showcase for reporting. Scaling should not happen too early. Projects should be used to reveal limitations and errors while they are still cheap. They should be linked to analytics and monitoring of effects. This is why protected sandboxes are needed — regimes in which one can test without destructive consequences — together with modern systems for evaluating effects that show not “how many events were held,” but what actually changed in the lives of children and teachers.

One more principle runs through the whole framework: limited normativity. In a world where there are no tested models and everything changes quickly, excessive regulatory rigidity is dangerous: it fixes mistakes and deprives the system of flexibility. Decisions should therefore be framework-based rather than directive, with the possibility of revision and continuous dialogue with all parties.

For the framework not to remain abstract, it is useful to see the whole field of work. If the proposed projects are brought together, they align around three axes and form nine semantic groups — a convenient map for seeing where the field is dense and where it is empty.

The first axis is the trajectory of the child’s own development: from teacher training and the culture of students’ AI use in school to higher education, the labor market, and integration into the global space. The second is school education itself: teacher preparation, principal preparation, and formation of a literate culture of AI use among children. The third is the characteristics of development: reducing social gaps, personalizing education, and protecting national foundations. At the center of this map are the two main actors that must meet: state policy and the offer of the EdTech market.

The value of such a map is that it shows gaps without mercy. Today, not all of these nine groups are covered either by state policy or by market supply. In some areas the state is active but the market is silent; in others the reverse is true. Some directions — especially protection of national foundations and reduction of the gap between city and village — risk remaining no one’s responsibility. These blank spots are the priority zones for action: not where work is already boiling, but where it is still absent although critically needed.

Let us return to where we began. 2026 has been declared the Year of Digitalization and Artificial Intelligence in Kazakhstan. A great deal can indeed be achieved during this year — but only if we correctly understand what exactly needs to be achieved.

The main temptation is to report implementation: launch platforms, distribute tools, show coverage. But we have gone through enough in this series to understand that the race for implementation without a governance framework leads exactly where others have already found themselves — to failures, rollbacks, and undermined trust. The real task of the Year of AI is not the number of launched projects, but the quality of the architecture built.

The good news is that the window is open. No one has a ready-made model, which means Kazakhstan is not in the role of a laggard trying to catch up. It is in the same conditions as the rest of the world. And it has a rare chance: not to copy someone else’s technology, but to build from the start an intelligent, governable, and human-centered model of school. A model in which AI is an assistant, not a replacement for the teacher; where human development takes priority over digital efficiency; where national identity is built into the foundation; where experiments are protected, and so are the vulnerable; and where the state maintains balance rather than driving volume.

Artificial intelligence is already in school. This is a given, and it is too late to argue with it. But what it becomes there — medicine or poison, an ally of the teacher or their gravedigger, a tool for development or a crutch for dependence — is not decided by the technology itself. It is decided by us. Today, at the start, while the window is open.

The main question of the era was and remains not “can AI be introduced into school.” It can. The question is how and why. And now we have an answer.

The material was prepared based on the results of foresight sessions held in autumn 2025 and winter 2026.