The effects of artificial intelligence on management education

Synthesis: Csaszar, Jacobides, and Zemsky treat generative AI as a strategic shock to the business school itself. Because large language models now perform well on the assignments that certify management competence, the authors ask whether the sector is entering a third epoch, after the applied era of the early 1900s and the research-driven era that followed the 1959 Gordon and Howell report. Applying a value-based framework to willingness to pay and delivery cost, they trace how AI could substitute for or complement Business Education, then set out teaching, research, and business model responses. Their conclusion is organizational rather than technological: incremental adaptation will not be enough.

Key Findings

1. Epoch framing, offered as a possibility. ChatGPT's 2022 release may mark a third epoch in management education, but the authors say the trajectory may also fail to transform the sector, as earlier threats from for-profit post-secondary education and MOOCs did.
2. A value-based lens. They apply willingness to pay and delivery cost, extending the analysis from students to faculty, employers, and donors, with the MBA as the flagship case.
3. Demand could move either way. AI may substitute for training as LLMs match student work and consulting firms anticipate 40% productivity gains for entry-level consultants, or complement it by raising the value of framework fluency, judgment, ethics, and leadership. The authors call the net effect an empirical question.
4. Teaching is unsettled at all three questions. Content must add AI as a business context and collaboration with AI analysts; method is stressed by assessment and participation; audience may shift from future consultants toward industry managers and mid-career executives.
5. Research is the moderating variable. AI can accelerate literature review, hypothesis generation, and synthetic data, and the authors argue that research capability decides whether AI complements or substitutes for business schools.
6. Competition and product mix will shift. Low-cost conversational learning lets edtech platforms and corporate learning attack from below, while undergraduate programs look more resilient than the MBA and executive education may grow first.
7. The business model is the binding constraint. Teaching revenue funds research, so shorter programs, price pressure, or falling demand create a structural deficit; three responses follow, and faculty incentives, accreditation, and weak managerial experience are the barriers.

A value-based lens on the business school

The paper is a conceptual strategy analysis, not an empirical study. It takes a framework built for firms, the value-based approach of Brandenburger and Stuart, and applies it to a sector that mostly does not compete on price. Value is set by what students and other stakeholders will pay and by the cost of the activities that deliver it. The authors extend the stakeholder set to faculty, employers, and donors, with the MBA as the flagship product.

The history gives the framework its stakes. The first epoch was applied and practitioner-led. The second began with the Gordon and Howell report of 1959 and made research the source of legitimacy, installing the faculty reward system the authors later identify as an obstacle. A third epoch would not simply add AI to the curriculum; it would change what the school sells, how it produces it, and who pays. The authors are careful: direction, magnitude, and timing are uncertain, and they present their contribution as a map of uncertainties and a menu of choices, not a forecast.

They note that Higher Education has absorbed earlier disruption claims and that MOOCs did not unseat business schools. Against this they set the possibility that AI differs because it is a general-purpose technology that now performs the analytical work schools teach.

Demand: substitute, complement, or both

MBA demand is derived demand. Students pay because employers pay a premium, and that premium rests on more than classroom learning: screening, peer networks, alumni access, reputation, and the residential experience all contribute. A technology that replaces analytical skill therefore need not replace the screening and networking functions, and one that improves learning need not protect the tuition price.

The substitution case is direct. LLMs score well on business school assignments and can take over tasks graduates once did. The authors cite consulting firms expecting 40% productivity gains for entry-level consultants. Because professional service firms sustain the MBA's return on investment, a fall in their hiring would feed back into demand. They read the drift toward shorter MBAs as early evidence of this pressure.

The complement case rests on a different claim about the degree's purpose. Extracting value from an AI analyst requires enough understanding to request, check, and revise its output, so framework knowledge becomes more valuable, not less. The authors expect the customer mix to shift: fewer future consultants, more industry managers who implement, and more small and mid-sized firms that could not previously afford the talent. Judgment, ethics, and change management become more valuable as analysis is automated, favoring mid-career managers over young generalists.

They do not resolve the two cases; which dominates depends on capability, adoption speed, and employer behavior.

Teaching, research, and the revenue that links them

The teaching analysis asks three questions. On content, the minimum is to treat AI as a business context, as earlier curricula treated globalization. A deeper move is to teach students to work with AI analysts and to design frameworks for AI-augmented decisions. The hardest is to strengthen critical thinking and judgment, including the limits the authors catalog: brittleness in unfamiliar situations, weak causal reasoning, hidden bias, opacity, adversarial inputs, and concept drift. Students should be able to audit and contest output, not merely consume it.

On method, assessment is already strained, and even class participation is hard to preserve when students can consult a model in real time. The lecture looks particularly exposed to automation, and AI tutoring plus personalized learning could change the mix of lecture, group work, and individual study. On audience, the authors expect a shift toward industry roles, continuous learning, and executive formats, partly because non-degree programs are easier to redesign.

Execution is where the paper is most concrete. Curriculum redesign collides with faculty incentives: promotion rewards research, and course changes tend to depress evaluations while they are refined. Some subjects are easier to automate than others, which makes reallocation politically harder. Accreditors require contact hours and have standards built for traditional delivery. Most faculty have little practice at pedagogical innovation, because teaching methods, unlike research methods, are not developed through doctoral training and peer review.

The research argument runs in parallel. AI lowers the cost of literature review and drafting, can suggest hypotheses, and can generate synthetic data when real data are scarce. The authors want management research to become more prescriptive and evidence-based, and they see a direct teaching payoff: schools that produce the research also hold the content. Academic freedom limits what leaders can direct, and publication norms are enforced by the field rather than any central authority. Doctoral programs are the lever they identify for long-run change.

The two agendas meet in the business model. Teaching generates most operating income and funds research, so anything that shortens programs, cuts price, or reduces enrollment threatens the research base. The authors offer three responses: hold tuition and differentiate on quality and experience, expand fundraising, or commercialize research. All three depend on the research transformation succeeding. Admissions, career services, and advancement can be automated, and deans, who rarely arrive with management experience, may be the least prepared for the organizational work.

Where the case method appears, and where it does not

The article does not discuss case-based teaching. This matters because a reader might assume otherwise. The case method is not one of its topics, its recommendations, or its examples of pedagogical change.

Case studies appear twice, incidentally. In Table 1, a "Custom Content Creator" is said to generate tailored materials, practice problems, and case studies. In the account of past teaching changes, the authors write that some case studies were replaced by business simulations and some readings by videos watched before class, and they call these modular changes with limited impact. Neither passage makes a claim about case discussion, instructor questioning, case writing, or how AI alters the preparation and conduct of a case class.

Two nearby claims bear on case teaching without addressing it. Class participation becomes problematic when students have real-time access to AI-generated insight, which applies to any discussion-based class. A virtual business simulator and dynamic simulations for risk-free experiential learning also appear among AI-enabled teaching innovations, making simulation the one case-adjacent pedagogy the paper actually treats. Readers should not cite this article for a case-method argument; it does not contain one.

What this means for practice

• Instructors. Audit what each assignment actually certifies. If a model can complete it, the assessment is measuring the model, not the student; shift weight toward judgment, verification, and revision of AI output.
• Program directors and curriculum designers. Treat AI as content and method, but cost the change against accreditation contact-hour rules and faculty evaluation cycles.
• Deans and administrators. Model the teaching-revenue dependency first. Shorter or cheaper programs are a business model decision, and the paper gives three financing paths to weigh.
• Researchers. Treat the school's own AI adoption as a research site. The authors make research quality the variable that decides whether AI complements the institution, and doctoral training the lever for changing it.
• Instructors who teach by case. Do not read this paper as guidance on case teaching; its transferable claims concern participation and assessment integrity, both stressed by AI.

Limitations

• The paper is conceptual. It offers no data, no systematic review, and no test of the substitution and complement mechanisms; the authors call the net demand effect an empirical matter.
• The framework is deliberately stylized and built around the MBA and elite schools in mature markets. Undergraduate, executive, and non-degree segments are discussed, but the competitive analysis is thinner there.
• The strategic claims depend on unsettled conditions: AI capability, regulation, and the speed of institutional adaptation. The authors decline to forecast and present scenarios instead, so the recommendations are contingent by design.
• The demand evidence is early. Consulting productivity expectations and edtech examples stand in for measured labor-market outcomes, and the paper offers no employer-side evidence of substitution beyond stated expectations.
• One author, Zemsky, is seconded to lead an AI corporate edtech venture, disclosed in his biography; the paper's declaration states no conflict of interest.

Citation

Csaszar, F. A., Jacobides, M. G., & Zemsky, P. (2026). The effects of artificial intelligence on management education. Strategic Organization, 24(2), 440-462.