Dr Guy W. Bate
Many features of education that appear pedagogically intentional are also responses to practical limits. A teacher can prepare only so many examples, offer only so many pathways through a concept and provide detailed feedback on only so much student work. Assessments capture learning at particular moments because continuously interpreting every learner’s progress has never been feasible.
Education technology has inherited many of these limits. Courses are commonly built around a fixed sequence of content. Feedback is concentrated at a few assessment points. Learner dashboards reduce progress to the indicators that a system can readily capture and display. These approaches provide necessary structure, but they also reflect the time, attention and processing capacity available to educators and institutions.
In a recent Harvard Business Review article, AI Is Revolutionizing Strategic Decision-Making, Felipe Csaszar (2026) describes how AI can relax similar constraints in strategy. He calls this unbounding. AI allows organisations to explore more options, build richer representations of their environment and test proposals against a wider range of perspectives.
Csaszar’s argument raises a useful question for education technology: which features of educational design exist because they support learning, and which persist because educators and systems have lacked the capacity to do anything else?
Expanding the educational possibility space
The first constraint Csaszar identifies is search. A strategy team can develop and assess only a small number of options before time runs out. AI lowers the cost of exploring a much larger possibility space.
Educational design faces a similar constraint. An educator introducing a difficult concept may prepare one explanation, one example and one activity. A course designer may create a single route through the material because developing several credible alternatives would take too long. Students who do not connect with that route must adapt to it or seek help elsewhere.
AI can expand the range of possibilities available before an educator decides what to use. It can help generate explanations at different levels of abstraction, examples drawn from different settings, alternative sequences of practice and activities that expose different aspects of a concept. An EdTech product could use the same capability to help educators anticipate common misconceptions or compare several ways of introducing difficult material.
The value lies in the range from which the educator can choose. Generating ten explanations does not mean presenting all ten to a learner. Too many routes can create confusion and weaken the coherence of a course. The educational opportunity is to explore more possibilities during design, then select and sequence those that best support the learner, the curriculum and the context.
This changes the role of personalisation. Instead of treating personalisation as the automated delivery of more content, EdTech providers can support educators in making better choices about which explanation, example or activity should come next. AI expands the search. Educational judgement gives the resulting pathway direction.
Seeing learning in higher resolution
Csaszar’s second constraint concerns representation. Organisations simplify complicated markets and competitive environments into frameworks, spreadsheets and periodic reports because these are manageable. AI can combine more variables and update them more frequently, creating what he describes as a richer, living representation.
Education also relies on simplified representations. A grade summarises performance on a particular task. A dashboard might show completion, time spent and quiz results. These indicators can be useful, but they reveal only part of what a learner understands. They may miss a misconception that appears across several activities, an improvement between early and later drafts, or a pattern in the questions a learner chooses to ask.
AI may allow learning systems to work with a broader range of evidence. A system could identify recurring reasoning errors across different tasks, distinguish a momentary mistake from a persistent misconception, or show how a learner’s explanation of a concept has developed over time. Educators could receive a more current picture of where support is needed rather than waiting for the next formal assessment.
Higher resolution, however, should not be confused with complete understanding. A model can detect patterns in the evidence available to it. It does not know the whole learner, their circumstances or the cultural meaning of what it observes. More data can create a stronger appearance of certainty even when important parts of the learner’s experience remain invisible.
For EdTech providers, richer representations therefore bring greater responsibility. Learners and educators need to know what evidence a system uses, what conclusions it draws and where uncertainty remains. They should be able to question or correct the picture presented. Privacy and agency become part of the educational design, especially when systems combine information across activities or over time.
A living model of learning should support a conversation about progress. It should not become an unquestionable account of the learner.
Making intellectual challenge more available
The third constraint in Csaszar’s argument is the difficulty of bringing multiple perspectives into a decision. Organisational discussions are shaped by hierarchy, confidence, politics and time. AI can support structured critique by taking on different roles, testing assumptions and simulating how others might respond.
This idea has considerable educational potential. Learners do not always have access to someone who can examine an argument, expose a weak assumption or offer a counterexample at the moment it would be useful. Teachers cannot sustain an individual dialogue with every learner throughout every activity.
AI can make this kind of intellectual scrutiny more consistently available. A learner developing a proposal might ask an AI system to identify unsupported claims, examine the proposal from a stakeholder’s perspective or construct the strongest case against it. A student working through a scientific explanation could be asked to respond to a plausible misconception. A learner preparing for professional practice could explore how different choices might affect a client, colleague or community.
The purpose is to deepen the learner’s reasoning. If the system supplies the argument, the response and the final evaluation, it may leave little intellectual work for the learner. Effective design keeps the learner responsible for interpreting the critique, deciding what is relevant and revising their position.
Synthetic perspectives also require careful framing. An AI-generated account of how a Māori community, disabled learner or cultural group might respond cannot replace engagement with people from that community. It may help reveal questions that designers and learners should investigate, but it cannot authorise an answer on their behalf. Co-design and participation remain essential.
Choosing educational boundaries deliberately
Csaszar is careful to explain that unbounding does not remove limits. It pushes some limits outward. This distinction becomes particularly important in education because boundaries can serve a pedagogical purpose.
Learners have finite attention. They need coherent sequences, opportunities to practise and time to struggle with difficult ideas. More explanations can obscure the underlying concept. More feedback can make learners dependent on external direction. Continually adapting a pathway can prevent a learner from developing persistence or seeing how separate ideas connect.
The question for educators and EdTech providers is therefore not how to remove every boundary. It is how to distinguish boundaries inherited from limited capacity from those deliberately chosen to support learning.
That distinction can guide product development. If AI allows a platform to generate many possible activities, who decides which activity serves the learning outcome? If a system can continuously interpret learner behaviour, which information is educationally useful and which should remain uncollected? If AI can supply constant critique, when should it remain silent so that a learner must judge their own work?
These decisions cannot be delegated entirely to the model. They require pedagogical knowledge, understanding of learners and attention to the social and cultural context in which education occurs.
Designing beyond scarcity
Csaszar’s idea of unbounding offers EdTech a way to think beyond incremental efficiency. The opportunity extends across three parts of educational design: considering a wider range of learning pathways, developing richer and more current representations of learning, and making intellectual scrutiny more consistently available.
Each expansion creates a corresponding responsibility. A wider range of pathways requires careful selection. A richer picture of learning requires transparency, privacy and room for learners to contest it. More frequent critique requires decisions about when intervention supports thinking and when it begins to replace it.
For EdTech providers, the starting point is to identify the practical limit a product is intended to relax. What could an educator, learner or institution now see, consider or explore that was previously out of reach? The next question is equally important: what educational boundary should remain, and who is responsible for setting it?
AI can expand the capacity available for educational design. The opportunity is to separate limits inherited from scarcity from boundaries chosen because they support learning. As those practical limits move outward, human judgement becomes more visible, and more consequential, in deciding what education should become.
References
Csaszar, F. A. (2026, September–October). AI is revolutionizing strategic decision-making. Harvard Business Review. https://hbr.org/2026/09/ai-is-revolutionizing-strategic-decision-making
————————————————
Dr Guy W. Bate is Thematic Lead for Artificial Intelligence, Director of the Master of Business Development (MBusDev) programme, and Professional Teaching Fellow in Strategy and Innovation at the University of Auckland Business School. He chairs the AI in Education Technology Stewardship Group for EdTechNZ, and has two decades of international industry experience in health technology, biotechnology and pharmaceuticals.