From “seat time” to skills mastery: redesigning the university for the age of AI

Prof. Mihnea Moldoveanu
Director of the Desautels Centre for Integrative Thinking
Founder of RotmanDigital, Rotman School of Management, University of Toronto

For decades, educators have known that one-to-one tutoring is among the most effective ways to help students learn. In 1984, educational psychologist Benjamin Bloom described what became known as the “2 Sigma Problem”: students who received individualized tutoring consistently outperformed those taught in conventional classrooms.

The challenge was never pedagogical. It was economic. Individual tutoring worked, but it was too expensive to provide at scale.

That constraint is beginning to disappear.

Recent advances in AI, particularly large language models and agentic AI systems, have made it possible to provide personalized guidance, feedback, and coaching to large numbers of students simultaneously. While these systems are not a substitute for faculty, they dramatically reduce the cost of interaction and support. As a result, they challenge some of the assumptions on which the modern university was built.

Most universities still operate according to a model developed for an era of scarcity. Expert explanation is scarce, so students are taught in cohorts. Personalized feedback is expensive, so assignments are graded selectively. Interaction is limited, so learning is standardized. Academic progress is measured largely by time spent in courses rather than by demonstrated mastery.

AI calls those assumptions into question.

The question facing higher education is not whether universities should adopt AI tools. Many already have. The more important question is whether AI enables a fundamentally different model of learning,: one organized around competency, dialogue, and continuous feedback rather than lectures, semesters, and credit hours.

From seat time to demonstrated competence

Most universities still define progress by time. Students enroll in courses, complete assignments, sit examinations, and accumulate credits until they qualify for a degree.

Competency-based education takes a different approach. Instead of measuring learning by hours spent in a classroom, it measures what students can actually do. Students advance when they demonstrate mastery of specific skills or knowledge, regardless of how long the process takes. The concept is not new. What has been missing is an affordable way to support learners individually as they progress at different speeds. Agentic AI changes that equation.

Properly designed AI tutors can answer questions, provide feedback, identify misconceptions, recommend learning activities, and adapt instruction to individual needs. Research increasingly suggests that these systems can improve learning outcomes when they are designed around sound educational principles rather than simply functioning as answer machines.

A recent randomized trial in a large Harvard physics course found that students working with a carefully designed AI tutor learned significantly more in less time than students in an active-learning classroom. Other studies have reported similar benefits when AI functions as a scaffold that guides learning rather than replacing it — a critical distinction, as unstructured use of AI can undermine learning by encouraging students to outsource thinking.

Structured use, by contrast, can deepen learning by providing frequent feedback, targeted coaching, and opportunities for practice. For universities, the goal should not be to deploy chatbots, but to create educational systems built around effective tutoring.

The assessment problem

The case for redesigning higher education is strengthened by a second development: the growing difficulty of assessing student learning using traditional methods.

Much of modern assessment relies on written work completed outside the classroom. Essays, reports, and take-home assignments have long served as proxies for understanding. Generative AI makes those proxies increasingly unreliable.

A widely discussed study from the University of Reading found that AI-generated exam answers frequently passed through standard grading processes undetected and often received higher grades than work produced by students. The details of any single study can be debated, but the broader trend is hard to ignore. Universities can no longer assume that a written submission reliably demonstrates individual competence.

This does not mean that learning has become impossible to assess. It means that assessment must evolve.

Increasingly, universities are experimenting with approaches that emphasize live performance and authentic demonstration. Oral examinations, project defenses, simulations, supervised problem-solving exercises, and workplace-style assessments all provide stronger evidence of what students actually know and can do.

These approaches are not new. Oral examinations, for example, have a long history in higher education. What is changing is the economics. AI systems can help conduct low-stakes practice sessions, provide feedback, and support assessment processes that would otherwise be too labor-intensive to scale. Human judgment remains essential, particularly in high-stakes decisions. But AI can make richer forms of assessment more practical than they have been in the past.

The dialogical university

If personalized tutoring and continuous assessment become widely available, universities may need more than new tools. They may need a new operating model.

Today's learning management systems were designed primarily to distribute content, collect assignments, and record grades. They function more like filing cabinets than learning environments.

An AI-native university would require something different: a platform built around continuous interaction.

Instead of treating learning as a sequence of lectures followed by periodic examinations, such a system would support ongoing dialogue among students, faculty, and AI agents. Learners would receive feedback as they work. Difficulties would be identified earlier. Instruction could adapt to individual needs and learning pace.

In this model, the university becomes less of a broadcasting system and more of a conversation.

The implications for teaching are significant.

When explanations, examples, and practice opportunities are available on demand, the most valuable contribution faculty can make is not repeated content delivery. It is designing learning experiences, creating challenges, mentoring students, supervising assessment, and helping learners make sense of complex ideas. The role of the professor shifts from transmitter of information to architect of learning.

That change should not be interpreted as diminishing the importance of faculty. If anything, it makes their educational expertise more central. Designing effective prompts, simulations, assessments, and interventions may become as important as delivering lectures once was.

What might change?

An AI-enabled university would not emerge overnight, nor would every institution adopt the same model. Yet several broad changes seem plausible.

• Academic records could focus more explicitly on demonstrated competencies rather than course completion.
• Students might progress at different rates, advancing when they demonstrate mastery rather than waiting for semester boundaries.
• AI learning coaches could provide continuous support, helping students navigate coursework, identify weaknesses, and access resources when needed.
• Assessment could rely less heavily on traditional essays and more on oral examinations, project work, live demonstrations, and other forms of authentic performance.
• Physical campuses might place greater emphasis on laboratories, studios, clinics, collaborative projects, and other activities that benefit from in-person interaction.

The common theme is a shift away from measuring educational activity toward measuring educational achievement.

Proceed with caution

None of this is risk-free. Poorly designed AI systems can encourage dependency rather than learning. Students may become overly reliant on tools that perform cognitive work for them. Bias, accessibility, privacy, and governance concerns remain real and require ongoing attention.

Public universities also operate within complex regulatory environments. Accreditation systems, funding mechanisms, faculty contracts, and governance structures were not designed for competency-based progression or AI-supported learning. Institutional change often lags behind technological change.

Nor should universities abandon core academic values in the pursuit of efficiency. Academic freedom, intellectual independence, scholarly inquiry, and rigorous standards remain essential. The goal is not to replace human judgment with algorithms, but to create institutions that use technology to support better learning.

The challenge is therefore not one of automation, but of redesign.

A different kind of university

Universities have repeatedly adapted to technological change. The printing press, mass literacy, broadcast media, personal computers, and the internet all altered how knowledge is created and shared.

AI may represent another such moment.

The most important consequence of AI in higher education may not be the automation of administrative tasks or the generation of course materials. It may be the removal of a longstanding educational constraint: the inability to provide personalized instruction to large numbers of learners.

If that constraint truly disappears, universities will face a choice. They can continue to organize learning around lectures, semesters, and credit hours, using AI merely to make existing processes more efficient. Or they can rethink the structure of higher education itself.

The argument for change is simple. Universities were designed for a world in which individualized tutoring was scarce and expensive. AI makes personalized learning increasingly abundant.

When the underlying economics of learning change, institutions eventually change as well. The task now is to decide what the next version of the university should look like.