1. Introduction and problem

In the direction of a kindergarten, a preschool, a CENDI or an infant school, a package of three gestures has become naturalised. The first is contractual: acquiring a licence. The second is organisational: appointing an educator, a coordinator or a deputy as “AI lead.” The third is communicative: circulating an internal memo that authorises, limits or celebrates the use of tools. All three are visible, auditable and cheap in leadership time. They allow the centre to exhibit, before inspectors, the network or families, that it is “already doing something” with artificial intelligence. The enunciative leap is enormous: one moves from an act of procurement, designation or drafting to the claim that pedagogical leadership is in place. That leap is not authorised by the evidence on school leadership or on early childhood education.

The thesis of this article is restrictive. Buying a licence, naming an AI lead or circulating an internal memo does not constitute pedagogical leadership. Leadership in early childhood centres is played out in another class of decisions: protecting play time as an organisational resource of the day; protecting educators’ judgement when a model generates programmes, plans or observation reports that look ready; and protecting the rights of children aged three to six, who are not autonomous users of generative platforms. Confusing the deployment of a tool with that protection is a category error. Wang (2021) reminds us that an educational leader’s decision is not only analytical: it is also moral. Fullan, Azorín, Harris and Jones (2024) warn that AI may ease the administrative load of leadership or, conversely, erode or replace leadership functions. What those three gestures do not do, by themselves, is decide on which side the kindergarten stands.

The problem is sharper in early childhood education for a professional reason that generic “AI for schools” guides do not show. The object of direction is not a subject timetable or a ranking of results. It is process quality: children’s everyday interactions with adults, peers, materials and space (OECD, 2021). TALIS Starting Strong 2024 shows that most leaders collaborate more than once a week with staff to improve how children play together and to observe staff practices and interactions, whereas developing or communicating a vision usually occurs at most once a month (OECD, 2025). That is the real work of direction. A memo, an AI lead without a pedagogical mandate, or a licence does not replace it.

This paper does not recycle UNESCO’s in-service teacher competence framework, nor the axes already treated in this series. The question is one of direction: what counts as pedagogical leadership when an early childhood centre “does AI.” The contributions are three: reconstructing the state of the art; examining three families of cases — adoption by principals, generative-AI use by middle leaders, and centre- and system-level policy —; and offering four tests to decide when a direction can claim that it leads, and not merely that it has bought, appointed or communicated.

2. State of the art: from pedagogical leadership to deployment leadership

Four strata that the literature and centre practice tend to mix should be kept apart. The first is pedagogical leadership in early childhood education, understood as shared responsibility for the quality of pedagogy, not as a post. The second is school leadership mediated by artificial intelligence: what principals and middle leaders do with AI systems or generative models. The third is the centre-policy gesture — licence, AI lead, memo — that claims to stand for the second and, often, for the first. The fourth is the early childhood rights frame, which sets an age threshold and a duty of human oversight that no tool deployment can elude.

In the pedagogical-leadership stratum, Heikka, Pitkäniemi, Kettukangas and Hyttinen (2021) show, with 130 professionals in Finnish centres — directors, teachers and care staff — and six interviews, that centres had adopted approaches congruent with distributed pedagogical leadership, and that implementation of distributed forms related positively to teachers’ capacity to lead reflection and learning in their teams. Teachers’ commitment was higher when that distribution was enacted sufficiently. Fonsén, Szecsi, Kupila, Liinamaa, Halpern and Repo (2023) compare directors’ and teachers’ discourses in Finland and Florida (in Finland, 13 directors and 15 teachers): there is a similar conceptualisation of distributed pedagogical leadership, with differences in the independence expected of teachers and in the extension of practices beyond the centre. Fonsén, Ahtiainen and Heikkinen (2023), with five focus groups of 15 Finnish leaders, find “pedagogical lenses” and the curriculum as a strategic tool: leadership is the means to interpret and implement the curriculum together with teachers. Fonsén, Marchant and Ruohola (2023) contrast Finland and Singapore: Finnish discourses are articulated as pedagogically focused leadership. In all of these studies, pedagogical leadership is not a technological annex: it is the everyday work of looking at classroom pedagogy.

TALIS Starting Strong 2024 saturates that portrait at system scale. Strong leadership — including distributed leadership and support for staff pedagogical learning — is associated with staff satisfaction with leadership, which in turn contributes to practices with children. Pedagogical responsibilities are more often shared with staff; administrative and financial ones, with external actors. Engagement in standard pedagogical-leadership practices is widespread; engagement in more strategic ones, less so (OECD, 2025). Starting Strong VI had anchored the problem in process quality: what counts is everyday interaction, not the regulation (OECD, 2021). Inference, marked as such: a centre that “leads on AI” and does not protect those interactions is not leading pedagogy. It is leading a tools file.

In the school-leadership-and-AI stratum, the evidence is recent and scarce. Arar, Tlili and Salha (2024) find no education-administration journals among the ten main platforms mentioning educational leadership and AI in Scopus. Wang (2021) synthesises a symbiotic role: AI can bring analytical efficiency to data-driven decisions, but that path may collide with value-based moral decision-making; the best is a blend, with AI as an extended brain and human judgement guided by values as a corrector. Adams and Thompson (2025) note that the literature is “extremely scarce” and project that models such as ChatGPT may streamline administration and communication. Fullan et al. (2024) pose the fork: the technology may lift mechanical loads or erode leadership functions; what the leader values remains under their control. None of those texts claims that a licence, an AI lead or a memo closes the problem.

In the rights stratum, UNESCO’s Recommendation anchors human rights and human oversight (UNESCO, 2021). The Guidance for generative AI requires a human-centred approach, pedagogical validation and an age threshold for independent conversations with generative platforms (Miao & Holmes, 2023). UNICEF (2021) starts from the Convention on the Rights of the Child and requires support for children’s development and prioritisation of the best interests of the child. The European Commission (2022) recalls that AI is already in schools. The U.S. Department of Education insists on humans in the loop and on not substituting the teacher (U.S. Department of Education, 2023). The OECD Digital Education Outlook 2023 locates tools and competences in the system ecosystem, not in a centre-level annex (OECD, 2023). Inference: the kindergarten cannot treat a four-year-old as a model user, nor the educator’s judgement as a cost to be optimised with a licence.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate a homogeneous effect size across principal interviews, middle-leader surveys and policy reports, but to articulate an argument about the direction of early childhood centres with verified sources. Inclusion criteria: (a) 2021–2026; (b) school direction, middle leaders, pedagogical leadership or ECEC centre leadership; (c) artificial intelligence, generative models or EdTech adopted by leaders; (d) relevance to kindergarten, preschool, CENDI or infant schools, or to the directional function those centres share with other stages; (e) peer-reviewed journal, DOI or UNESCO, OECD, UNICEF, European Commission or ministry report; (f) verifiable DOI or publisher page. Axes already used in this series, including initial teacher education, were excluded.

The search was executed on 25 August 2026 on DOI pages, Emerald, Frontiers, SAGE, Taylor & Francis, OECD iLibrary, UNESDOC, UNICEF Innocenti, GOV.UK, the European Commission and repositories (Jyväskylä, Helsinki, Melbourne). Each source was verified against at least one of those pages. The corpus was organised into adoption by leaders, generative-AI use by middle leaders, and centre- and system-level early childhood policy.

The analysis distinguished three enunciative statuses. Empirical finding: what was observed or measured in the leader sample or, in TALIS, in the ECEC leader population. Conceptual or normative frame: what a framework, a guide or a recommendation prescribes. Pedagogical inference: the translation this article proposes for the direction of kindergartens, preschools, CENDI and infant schools, marked as such. The limits are those of a narrative review: language bias towards English and a scarcity of Latin American studies of ECEC direction and AI (section 9).

4. Case 1. Adopting a system is not directing a kindergarten: seven leaders and the first generation of implementation

Tyson and Sauers (2021) published in the Journal of Educational Administration (vol. 59, no. 3, pp. 271–285) a qualitative case study of school leaders’ experiences with the adoption and implementation of artificial-intelligence systems. The purpose was to examine the factors that led educational administrators to adopt a concrete programme — ALEKS — and their perceptions of the implementation process. The design included structured interviews with seven people who had adopted AI programmes in their schools, identified by purposive and snowball sampling, in Georgia. Two themes emerge: leaders were actively engaged in conversations related to adoption and implementation, and they created organisational structures to ensure them. The authors underline that the literature on artificial intelligence and school leadership is extremely limited (Tyson & Sauers, 2021).

Status of the evidence. Empirical finding: in that small Georgia sample, with that product and at that time — before the public explosion of mass-consumer generative models — direction “does AI” mainly as an adoption conversation and as the design of organisational structures. It is not a finding of pedagogical leadership in early childhood education. The study does not observe rooms for three- to six-year-olds, does not measure process quality, does not evaluate play time or educators’ judgement, and does not ask whether the tool displaced interactions. The product was an already commercialised AI system, not a generative text model. Berkovich (2025) locates it, rightly, in what Rogers would call the innovators group, and declares it in part obsolete as a portrait of current use. Its value for this article is not the system. It is the leadership gesture it documents: talking about adoption and building structures. That is exactly what a licence and an AI lead claim to exhibit. Pedagogical inference, marked as such: a kindergarten director who cites this kind of implementation to claim “we already lead on AI because we adopted it and gave it a structure” makes the leap the thesis forbids. Adoption is an act of management. Pedagogical leadership in ECEC starts afterwards, or it does not start.

Conceptual saturation comes from Wang (2021): AI can extend the leader’s analytical capacity; it cannot replace moral decision. Adopting a system is a decision. It is not, by the fact of adopting it, a pedagogical decision about a four-year-old’s day. Adams and Thompson (2025) project management gains. Fullan et al. (2024) recall the inverse scenario: the erosion of leadership functions. Seven leaders who create implementation structures do not authorise a pedagogical victory in a CENDI.

5. Case 2. The middle leader generates the instructional artefact: 302 leaders and the ceiling of assisted AI

Berkovich (2025) published in Frontiers in Education (vol. 10, art. 1643023) an online survey, in January 2025, of 302 public-school leaders in Israel. Sampling was convenience-based. 80.8% of respondents were women. Levels were almost evenly split: 50.7% elementary and 49.3% secondary. By role, the table reports 135 subject heads, 55 department heads, 49 social-activities coordinators, 38 counsellors, 16 vice-principals and 9 principals. The instrument covered 19 tasks in five domains — managerial, instructional, social, political and moral — and asked, for each, whether the person used generative AI. The introduction listed examples (ChatGPT, Copilot, Claude, Gemini, Perplexity).

Diffusion findings, read with Rogers, place about 50% of the sample at the early-majority stage, about to cross into the late majority. The most frequent tasks were developing educational programmes for age groups or the whole school (60.3%), developing lesson plans or resources for staff (57.3%), planning professional-development workshops (49%), creating communication materials for staff and families (48.7%), and planning, improving or drafting observation reports for teacher evaluation (48%). The least frequent were budget planning (25.8%) and diversity, equity and inclusion issues (34.1%). The instructional domain obtained the highest mean use (0.50), above social (0.43), political (0.41), managerial (0.37) and moral (0.36). Novice teachers (0–6 years, n = 74) integrated generative AI more in the managerial domain than experienced ones (16 years or more, n = 98): M = 0.46 versus M = 0.31, t(170) = 2.76, p = 0.006. Berkovich concludes that AI-assisted instructional leadership is emerging, and warns of technological dependence and the need for a professional ethical identity, not only user skill (Berkovich, 2025).

Status of the evidence. Empirical finding: in that 2025 Israeli sample, not representative, generative AI is no longer an innovators’ experiment. It is used widely, and more to produce instructional artefacts — programmes, plans, observation reports — than to decide in the moral domain. The observation-report figure is the closest to kindergarten direction: almost half report using a model to plan, improve or draft those reports. It is not an early-childhood finding: the sample is elementary and secondary, with only nine principals. It is not a finding of output quality or of contrast with a real observation. Pedagogical inference, marked as such: if the middle leader of a kindergarten — cycle coordinator, deputy, reference educator — drafts with a model the room programme or the report of what they “saw,” the gesture saves writing and may colonise judgement. Pedagogical leadership in Heikka et al. (2021) and Fonsén et al. (2023) consists in teachers leading reflection on their team’s pedagogy. An observation report generated before anyone has been in the room inverts that sequence: it produces the text of the gaze without the gaze. That is not assisted instructional leadership. It is the substitution of the act of directing.

Normative saturation is clear. England’s Department for Education states that teachers, leaders and staff must use professional judgement; any content produced requires critical judgement; the quality of final documents remains the responsibility of the professional and the organisation, regardless of the tools; generative AI does not replace a human expert’s judgement; technology should not replace the teacher–pupil relationship (Department for Education, 2025). Miao and Holmes (2023) require pedagogical validation and an age threshold. UNICEF (2021) requires the best interests of the child. Inference: a programme “for the four-year-old group” generated by a middle leader is not, by existing, an act of pedagogical leadership. It is a draft whose admission or rejection is the act of leadership. If the centre celebrates the draft as evidence that “we already use AI,” it has inverted the criterion.

6. Case 3. The AI lead, the memo, and what ECEC leaders actually do each week

Kafa (2025) published in the International Journal of Educational Management (vol. 39, no. 8, pp. 98–115) a qualitative study with semi-structured interviews of 43 primary and secondary school leaders across five districts in Cyprus, a highly centralised system. Findings indicate that digital tools improve communication and administrative efficiency, and that challenges of poor infrastructure and insufficient targeted training persist. Leaders recognise AI’s potential to support their practice — administrative tasks, greater speed — and, at the same time, report limited knowledge, lack of appropriate training and the need for ongoing support. The portrait is consistent: in a system that does not delegate, the leader imagines AI as managerial relief rather than pedagogical redefinition, and asks for the training that the AI-lead gesture claims to simulate.

Status of the evidence. Empirical finding: in that Cypriot sample, AI enters direction as a promise of faster administration and as a knowledge-and-support deficit. It is not a kindergarten finding. It is not a finding that naming an AI lead closes the deficit. Pedagogical inference: the “AI lead” that appears in English early-adopter chronicles — Ofsted, from 21 interviews with schools, colleges and trusts with at least twelve months of use, describes AI champions as those who create a “buzz” and support staff (Department for Education, 2025) — is a diffusion device. It may be useful. It is not, by existing, pedagogical leadership of an early childhood centre. An internal memo is the same device in textual form. Kafa shows the gap those devices do not fill: knowledge, continuing training, a policy frame. Fullan et al. (2024) recall that leadership is human connection, and that it is hard to see how AI would replace that function. An AI lead who is not in the room, or a memo that does not change the timetable, does not protect play time or the judgement of those who are there.

The sectoral anchor is TALIS Starting Strong 2024 (OECD, 2025). There, leaders of early childhood education are present. The practices most leaders report more than once a week are collaborating with staff to improve how children play together and observing staff practices and interactions with children. Developing or communicating a vision for the setting occurs, for most, at most once a month. Distributed leadership is a common feature and is associated with greater staff satisfaction with leadership. Pedagogical responsibilities are shared more with staff; administrative and financial ones, more with external actors. In most systems a majority of leaders report needing more institutional support: those figures exceed 80% in pre-primary settings in Colombia, Japan, Morocco, Spain and Türkiye. Finland launched Towards Evolving Leadership (VEPO 2035), with pedagogical leadership as one of four domains. Ireland, in Nurturing Skills 2022–2028, provides for designated leadership roles focused on pedagogy and mentoring (OECD, 2025).

Status of the evidence. Empirical finding: the weekly work of those who direct an ECEC setting, in systems that take part in TALIS, is not deploying tools. It is looking at classroom pedagogy and joint play, in distribution with staff. The policies OECD highlights as capacity-building do not create “AI leads”: they create pedagogical-leadership frames and pedagogy and mentoring roles. Inference, marked as such: if weekly leadership time is in observation and joint play, every hour absorbed by managing a licence, by an AI lead or by memos is an hour not invested in that work, unless the tool is explicitly in the service of protecting it. The memo that “authorises ChatGPT for staff” does not observe a room. The AI lead who “answers prompt questions” does not, by that function, improve how children play together. Starting Strong VI had said that process quality is the most proximal driver of children’s development (OECD, 2021). Leadership that does not touch that proximity is not pedagogical, however digitally ostensible.

7. Inferential frame: four tests for claiming that direction “leads” on AI

The frame that follows is this article’s pedagogical inference, anchored in the cases and the verified instruments. It is not a new international standard. It distinguishes four tests. If a kindergarten, preschool, CENDI or infant-school direction does not pass them, it cannot declare that a licence, an AI lead or a memo constitutes pedagogical leadership.

7.1. Test of play time as an organisational resource, not a slogan. TALIS Starting Strong 2024 locates in leaders’ weekly practice collaboration to improve how children play together (OECD, 2025). Starting Strong VI locates everyday interactions as the engine of process quality (OECD, 2021). Inference: pedagogical leadership is verified in the timetable, in staffing and in what is interrupted. If adopting a tool cuts play time or fragments it without an explicit, reviewable didactic reason, direction has done deployment management, not leadership. Protecting that time is a directional decision. It does not appear in the licence contract or at the foot of the memo.

7.2. Test of educators’ judgement, not of artefact fluency. Heikka et al. (2021) link distributed pedagogical leadership to teachers’ capacity to lead reflection in their teams. Fonsén et al. (2023) place pedagogical lenses and joint interpretation of the curriculum at the centre of the craft. Berkovich (2025) shows that leaders’ most intense use of generative AI is the production of programmes, plans and observation reports. The DfE requires that the final document remain the professional’s responsibility (Department for Education, 2025). Wang (2021) requires that moral decision correct the analytical one. Inference: evidence of leadership is not a portfolio of generated outputs or the appointment of someone who reviews them nominally. It is the capacity, observed in the centre, to accept, reject or rewrite an artefact with classroom didactic reasons, and not to sign an observation report the model wrote. An AI lead who has no mandate to say “this does not enter the room” does not protect judgement: they administer it.

7.3. Test of young children’s rights, not of platform users. Miao and Holmes (2023) set an age threshold for independent use of generative platforms. UNICEF (2021) requires the best interests of the child and development. UNESCO (2021) requires human oversight. The DfE states that technology should not replace the teacher–pupil relationship and that there are more immediate benefits and fewer risks in teacher-facing use (Department for Education, 2025). Inference: an ECEC centre cannot treat a child aged three to six as an autonomous interlocutor of a model or as the addressee of generated content without situated adult mediation. The memo that “allows AI in the classroom” without that boundary fails the test by construction. Leadership is the boundary.

7.4. Test of pedagogical distribution, not of a technological silo. Heikka et al. (2021) and TALIS Starting Strong 2024 show that the leadership that counts is distributed in the team, associated with staff satisfaction and with practices with children (OECD, 2025). Fullan et al. (2024) and Kafa (2025) agree on a point of method: the technology is not sustained in a single post. Inference: an AI lead housed outside pedagogical coordination, without protected time in team meetings and without a classroom criterion, is a silo. A memo issued from direction without subsequent work with educators is a textual silo. The place of knowledge is the team that observes, interprets and decides. OECD (2023) locates the problem in the ecosystem. Cutting the response to three centre gestures is not answering.

The frame admits management-support tools if they free real time to be in the room (Fullan et al., 2024; Adams & Thompson, 2025); middle leaders who use models as drafts provided professional judgement closes the file (Department for Education, 2025; Wang, 2021); and AI leads if their mandate is protection of time, judgement and rights, not diffusion of the tool. It rejects declaring leadership by the invoice of a licence (Tyson & Sauers, 2021); by the organogram of a champion (Department for Education, 2025); by a memo; by the quantity of programmes or reports generated (Berkovich, 2025); and by treating young children as users of generative models (Miao & Holmes, 2023; UNICEF, 2021).

8. Discussion

Three tensions organise the discussion. The first is between adoption and direction. It is a finding that leaders who were already implementing AI talked about adoption and created structures (Tyson & Sauers, 2021), that 2025 middle leaders use generative models mainly in the instructional domain (Berkovich, 2025), and that in a centralised system AI is imagined as administrative relief in the face of a knowledge deficit (Kafa, 2025). It is a frame that the leader’s decision is also moral (Wang, 2021) and that AI may ease or erode leadership (Fullan et al., 2024). It is not a finding that adopting, generating or appointing equals directing a kindergarten. The policy of the three gestures measures what management knows how to measure — contracts, posts, texts — and declares what only situated pedagogical leadership would authorise.

The second is between the instructional artefact and the educator’s craft. Berkovich (2025) documents programmes, plans and observation reports as main uses. The DfE documents that the professional remains responsible for the final document (Department for Education, 2025). Heikka et al. (2021) and Fonsén et al. (2023) document that ECEC pedagogical leadership lives in team reflection and in the lenses with which the curriculum is read. A middle leader faster at producing text and equally absent from the room is not directing better: they are producing more file. In early childhood education, the file is not the place of pedagogy. The room is (OECD, 2021, 2025).

The third is between the diffusion lead and the leader of process quality. Ofsted describes champions who create a “buzz” (Department for Education, 2025). Kafa (2025) describes leaders who ask for continuing training. TALIS describes leaders who, each week, observe and intervene on joint play and, at most once a month, articulate a vision (OECD, 2025). The capacity policies OECD records — VEPO 2035, Nurturing Skills — invest in pedagogical leadership, not in the figure of the deployer. Fullan et al. (2024) ask that no assumptions be made about AI’s future and that energy go into social intelligence. In a kindergarten, that social intelligence is the team of educators and the relationship with young children. A memo does not create it. Arar et al. (2024) show that the academic field of educational leadership and AI still has no home in school-administration journals. That orphanhood is reproduced in centres when AI is treated as a purchasing problem.

9. Limits

This review is narrative. It does not apply PRISMA or estimate combined effects. The leadership-and-AI cases have biased geographies (United States, Israel, Cyprus, England) and uneven n (7, 302, 43; Ofsted, 21 interviews). Tyson and Sauers (2021) predate mass-consumer generative AI and do not study a kindergarten. Berkovich (2025) uses convenience sampling, mixes elementary and secondary, includes few principals and does not measure output quality. Kafa (2025) is primary and secondary in a centralised system. Transfer to CENDI, kindergartens and infant schools is inference, not sample equivalence. TALIS Starting Strong 2024 and Starting Strong VI are the sectoral empirical base and do not measure AI. Heikka et al. (2021) and Fonsén et al. (2023) are pedagogical leadership without AI. Wang (2021), Adams and Thompson (2025) and Fullan et al. (2024) are conceptual or positional. UNESCO, UNICEF, the European Commission and the DfE are prescriptive. No Latin American studies of ECEC direction and AI were located with the same degree of DOI; that absence is a gap. The inferences in section 7 are directional hypotheses, not implementation evidence.

10. Conclusions

Buying a licence, naming an AI lead or circulating an internal memo does not constitute pedagogical leadership in an early childhood centre. The verified evidence does not authorise that declaration. Seven leaders who adopted an AI system talked about adoption and created structures; that is implementation, not classroom pedagogy (Tyson & Sauers, 2021). Among 302 school leaders, generative-AI use concentrates on programmes, plans and observation reports, not on the moral domain (Berkovich, 2025). Among 43 leaders in a centralised system, AI appears as an administrative promise and a knowledge deficit (Kafa, 2025). In TALIS early childhood settings, weekly directional work is observing interactions and improving joint play; vision is monthly; the leadership associated with staff satisfaction is distributed (OECD, 2025). ECEC pedagogical leadership is enacted in teams, pedagogical lenses and curriculum (Heikka et al., 2021; Fonsén et al., 2023). The leader’s decision is also moral (Wang, 2021). AI may ease or erode the craft (Fullan et al., 2024). Current law and guidance require human oversight, pedagogical validation, an age threshold and the best interests of the child (UNESCO, 2021; Miao & Holmes, 2023; UNICEF, 2021; Department for Education, 2025).

Where the sources do not measure the direction of a kindergarten, this article does not affirm it. Where they measure adoption, artefacts and posts, it does not translate them into leadership. Directing an early childhood centre in the face of artificial intelligence is protecting play time as a resource of the day, educators’ judgement as the admission criterion of any artefact, and young children’s rights as the boundary of what is not deployed. The rest is management. It may be necessary. It is not enough, and it must not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Adams, D., & Thompson, P. (2025). Transforming school leadership with artificial intelligence: Applications, implications, and future directions. Leadership and Policy in Schools, 24(1), 77–89. https://doi.org/10.1080/15700763.2024.2411295
  2. Arar, K., Tlili, A., & Salha, S. (2024). Human-machine symbiosis in educational leadership in the era of artificial intelligence (AI): Where are we heading? Educational Management Administration & Leadership. https://doi.org/10.1177/17411432241292295
  3. Berkovich, I. (2025). The rise of AI-assisted instructional leadership: Empirical survey of generative AI integration in school leadership and management work. Frontiers in Education, 10, 1643023. https://doi.org/10.3389/feduc.2025.1643023
  4. Department for Education. (2025). Generative artificial intelligence (AI) in education. GOV.UK. https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education
  5. European Commission. (2022). Ethical guidelines on the use of artificial intelligence (AI) and data in teaching and learning for educators. Publications Office of the European Union. https://doi.org/10.2766/153756
  6. Fonsén, E., Ahtiainen, R., & Heikkinen, K.-M. (2023). Finnish early childhood education and care leaders’ conceptualisations and understandings of pedagogical leadership. In M. Modise, E. Fonsén, J. Heikka, N. Phatudi, M. Bøe, & T. Phala (Eds.), Global perspectives on leadership in early childhood education (pp. 11–26). Helsinki University Press. https://doi.org/10.33134/HUP-20-2
  7. Fonsén, E., Marchant, S., & Ruohola, V. (2023). Leadership in early childhood education: Cross-cultural case studies before and during the COVID-19 pandemic. Educational Management Administration & Leadership, 53(4), 850–869. https://doi.org/10.1177/17411432231194849
  8. Fonsén, E., Szecsi, T., Kupila, P., Liinamaa, T., Halpern, C., & Repo, M. (2023). Teachers’ pedagogical leadership in early childhood education. Educational Research, 65(1), 1–23. https://doi.org/10.1080/00131881.2022.2147855
  9. Fullan, M., Azorín, C., Harris, A., & Jones, M. (2024). Artificial intelligence and school leadership: Challenges, opportunities and implications. School Leadership & Management, 44(4), 339–346. https://doi.org/10.1080/13632434.2023.2246856
  10. Heikka, J., Pitkäniemi, H., Kettukangas, T., & Hyttinen, T. (2021). Distributed pedagogical leadership and teacher leadership in early childhood education contexts. International Journal of Leadership in Education, 24(3), 333–348. https://doi.org/10.1080/13603124.2019.1623923
  11. Kafa, A. (2025). Exploring integration aspects of school leadership in the context of digitalization and artificial intelligence. International Journal of Educational Management, 39(8), 98–115. https://doi.org/10.1108/IJEM-11-2024-0703
  12. Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  13. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  14. OECD. (2023). OECD Digital Education Outlook 2023: Towards an effective digital education ecosystem. OECD Publishing. https://doi.org/10.1787/c74f03de-en
  15. OECD. (2025). Results from TALIS Starting Strong 2024: Strengthening early childhood education and care. OECD Publishing. https://doi.org/10.1787/20af08c0-en
  16. Tyson, M. M., & Sauers, N. J. (2021). School leaders’ adoption and implementation of artificial intelligence. Journal of Educational Administration, 59(3), 271–285. https://doi.org/10.1108/JEA-10-2020-0221
  17. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  18. UNICEF. (2021). Policy guidance on AI for children 2.0. UNICEF Innocenti. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children
  19. U.S. Department of Education, Office of Educational Technology. (2023). Artificial intelligence and the future of teaching and learning: Insights and recommendations. U.S. Department of Education. https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
  20. Wang, Y. (2021). Artificial intelligence in educational leadership: A symbiotic role of human-artificial intelligence decision-making. Journal of Educational Administration, 59(3), 256–270. https://doi.org/10.1108/JEA-10-2020-0216