Two recent contributions to the discussion of AI and higher education appear, at first, to pull in opposite directions. In an interview with Lance Eaton, the computer scientist David Bachman describes a world in which autonomous and agentic systems are advancing so quickly that universities risk becoming dysfunctional unless they change substantially. In a Times Higher Education article, Jelena Belic and Kritika Maheshwari warn that reshaping universities around an untested AI economy would be a serious mistake. One urges urgency; the other counsels restraint.

The tension is real. But it need not end in a choice between technological complacency and technological surrender. Taken together, the two pieces suggest a more demanding position: universities must change, but they must not allow uncertain corporate forecasts about AI and work to dictate what education is for.

Bachman’s interview is compelling because it starts with a concrete curricular problem rather than with a generic call for “AI skills”. In his introductory computer-science course, he no longer assumes that writing code line by line will remain the central capability for graduates. His response is not simply to let students ask an AI to write programs. Instead, he proposes to move from syntax to structure. Students will solve programming problems through executable flowcharts, learning algorithmic thinking before using agentic coding tools. When those tools generate Python, students will work backwards: they will translate the code into a flowchart in order to demonstrate that they understand what the code is doing (Eaton, 2026).

There is an important educational principle here. If AI makes one surface skill less valuable, the right response is not necessarily to abandon the underlying intellectual practice. It may be to make that practice more visible. Bachman’s distinction between algorithmic coding and algorithmic thinking is useful well beyond computer science. In many subjects, the question is not whether students will use AI, but whether they can explain the reasoning, knowledge and judgement that should guide its use.

Bachman also resists the temptation to offer a universal formula. His proposal is deliberately course-specific. What makes sense in an introductory computer-science class will not necessarily make sense in calculus, machine learning, history or a hands-on sculpture studio. He recognises that the impact of AI will be uneven across disciplines. This is a significant qualification to his broader sense of urgency. The curriculum may need to change, but it should not all change in the same way or at the same speed (Eaton, 2026).

Belic and Maheshwari make the necessary counterargument. They caution against treating predictions about an “AI economy” as established fact and then remaking higher education around them. The future of work is not settled. AI may enhance some human work while eliminating other tasks and roles; the balance between those tendencies will depend on technology, business choices, public policy and regulation. To reorganise universities around particular assumptions about which AI skills employers will value in five years is, therefore, a high-risk strategy (Belic & Maheshwari, 2026).

Their deeper objection concerns the purpose of higher education. Students are not only future employees. They are also citizens, voters and moral agents. Critical thought, intellectual autonomy, communication and ethical judgement do not become less important because AI tools are becoming more capable. Indeed, the ability to assess AI-generated claims depends on learners already having knowledge, reasoning and independence of their own. These are are the key conditions under which AI can be used responsibly (Belic & Maheshwari, 2026).

This is where the two arguments meet. Bachman’s proposed AI centre rests on three pillars: understanding AI’s wider social, legal, environmental and ethical impacts; knowing how and when to use it; and understanding how it works. Belic and Maheshwari make a parallel case for a broad critical AI literacy that goes beyond tool proficiency to include questions of training data, bias, misinformation, privacy, environmental cost and the changing organisation of work. Neither position is satisfied with teaching students to prompt a particular product efficiently.

The disagreement is therefore not really about whether students should encounter AI. They should. The disagreement is about what should drive curricular change. The danger identified by Belic and Maheshwari is that universities might confuse education with vendor-led workforce preparation: a rolling sequence of short-lived training courses built around the current tools of a highly concentrated industry. The danger identified by Bachman is that universities might move so slowly that students leave without the conceptual, practical and critical resources to navigate technologies that are already reshaping some fields.

A sensible response is to distinguish between durable capabilitiesandcontingent tools. Universities should invest heavily in the former: disciplinary knowledge, critical reasoning, communication, creativity, ethical judgement, collaborative work and, where appropriate, algorithmic thinking. They should offer students structured opportunities to experiment with the latter: to use AI systems, test their claims, recognise their limitations and document their role in a piece of work. The goal is neither to ban AI nor to normalise uncritical dependence on it. It is to develop agency in a changing technical environment.

This also implies a different model of institutional change. Universities do need to move faster than traditional curriculum-review cycles often allow. But speed should mean building the capacity for careful experimentation: supporting staff, running course-level pilots, sharing evidence across disciplines, involving students in evaluation, and revising approaches in the light of what is learned. It should not mean imposing the same AI requirement on every programme because a technology company or a labour-market forecast says that the future has already arrived.

Bachman’s flowchart experiment is a useful example of this approach. It takes seriously the possibility that agentic coding will change professional practice, yet it does not assume that students should simply surrender their understanding to the agent. Belic and Maheshwari’s warning provides the complementary discipline: no experiment, and no forecast, should be allowed to redefine the university’s public purpose without evidence, democratic debate and attention to whose interests are being served.

The future direction of AI is genuinely unclear. That is precisely why universities should neither stand still nor rush to become training departments for an imagined economy. Their task is to help students understand and shape the technical, social and political worlds they inhabit. A curriculum fit for an uncertain AI future will be responsive enough to engage with new tools, but grounded enough not to mistake those tools for the purpose of education.

References

Belic, J., & Maheshwari, K. (2026, August 19). Reshaping universities to suit the untested AI economy is wrong. Times Higher Education. https://www.timeshighereducation.com/opinion/reshaping-universities-suit-untested-ai-economy-wrong

Eaton, L. (2026, August 25). “I don’t think the possibilities of autonomous agentic coding have really sunk into the general public’s vision of what AI is”: Part 2 of an interview with Professor David Bachman. AI + Education = Simplified. https://aiedusimplified.substack.com/p/i-dont-think-the-possibilities-of?utm_source=email&redirect=app-store-no-desktop&inbox=true&utm_campaign=email-read-in-app&triedRedirect=true