There is much to consider regarding university capstone subjects. When mentioned, responses typically fall into one of two categories: either an enthusiastic endorsement of their transformative potential or a weary acknowledgment of the overwhelming expectations placed upon them, expectations that no single course could realistically meet. Both reactions are understandable and reflect a long-standing tension surrounding capstone courses.
Many universities are currently redeveloping their capstone subject models. They are examining what these subjects entail, what is expected from them, and whether the established conditions enable students to achieve their objectives. The introduction of AI in higher education has heightened the urgency of addressing these questions, especially when it is used as a checkpoint of assurance of learning.
Why capstones matter: contested purposes and persistent expectations
The simplest definition of a capstone subject is a culminating experience that comes at or near the end of a degree, but there is healthy disagreement; a dominant view is that their purpose is synthesis: drawing together what students have already learned and integrating it into something more coherent and more demonstrably their own. This is the foundational claim, and it is broadly accepted. What is less settled is what that synthesis should look like, who it is for, and what it is supposed to prove.
Some disciplines frame capstones primarily as an employability mechanism, a final subject designed to signal professional readiness to future employers and to develop the skills that workplaces need: communication, project management, decision-making, and collaborative problem-solving. Others position capstones as a scholarly threshold, the moment where students are inducted into research cultures and expected to frame significant disciplinary questions in genuinely original ways. And others treat the capstone primarily as an institutional instrument, a site for assurance of learning, where the quality of whole-of-course outcomes can be confirmed and documented.
These purposes are not necessarily incompatible, but they do create genuine tension. A subject designed to signal employability may not look like one designed to demonstrate research capability. A subject designed to certify learning outcomes mechanically may not feel like one designed to develop the student as a whole person as they transition into professional life. When you load all these expectations into a single subject, the result is a curriculum that is asked to simultaneously integrate, authenticate, develop, signal, and certify, which is a substantial and complex task.
In practice, this means that before designing or evaluating a capstone, it is necessary to be explicit about what the subject is primarily positioned to do. That clarity is harder to achieve than it sounds, and it is critical in a context where AI has complicated every one of those purposes.
Positioning the capstone: culmination, transition, or both?
There are two broad positions a capstone can occupy in a degree, and they are not quite the same thing, even though they are often conflated.
The first is the capstone, the culmination. In this framing, the final subject is the moment where accumulated learning is formally demonstrated and certified. It looks backwards across the degree and asks: What has this student achieved? It is closely tied to course learning outcomes, accreditation requirements, and institutional claims about graduate quality. The capstone here functions as a kind of evidence locker, the place where you can point when someone asks whether your graduates meet the standard.
The second position is the capstone as a transition. Here, the subject looks forward rather than backward. It is less concerned with certifying past attainment than with preparing students for what comes next, whether that is employment, postgraduate study, or professional practice. The question it asks is not “what have you learned?” but “are you ready for what comes after this?” Transition-focused capstones often involve industry partnerships, community-engaged projects, research internships, or other activities that extend beyond the formal curriculum map.
In practice, most capstones are asked to do both. They are expected to look backwards and confirm attainment while simultaneously looking forward and demonstrating readiness. This dual positioning is entirely understandable from an institutional perspective, but it significantly raises the stakes of everything that happens in the subject. When a capstone serves both the culmination and transition functions, the assessment design must do an enormous amount of work. The tasks need to be rigorous enough to certify learning outcomes, authentic enough to demonstrate professional capability, open-ended enough to allow genuine student agency, and structured enough to produce comparable evidence across a diverse cohort.
Capstones as a high-risk, high-reward curriculum
Capstones are disproportionately consequential relative to their credit-point load. In most degree structures, they represent a single subject out of many, yet they carry pedagogical weight that far exceeds their formal proportion of the course. This is what makes them genuinely risky to design and run.
The risk is partly structural. Capstones typically involve high levels of student autonomy, open-ended tasks, and varied supervision arrangements. A student working on an industry project may have a very different experience from one completing an honours-equivalent thesis or a community-based research project. These are not equivalent assessment conditions, and the variability they introduce creates real challenges for consistency of judgement, comparability of grades, and the defensibility of outcome claims.
The risk is also reputational. Because capstones function as a proxy for whole-of-course learning, a poorly designed or inconsistently assessed capstone can undermine confidence in the entire degree. When accreditation bodies, employers, or government agencies ask for evidence of graduate capability, the capstone is often the most direct answer available. That answer needs to be credible.
At the same time, the open-endedness and autonomy that create these risks are also what make capstones genuinely valuable. Students who can frame their own research questions, manage complex projects independently, and produce work that reflects their authentic disciplinary identity tend to find the capstone the most meaningful part of their degree. The reward is real. The challenge is to build the conditions under which that reward is accessible to all students, rather than primarily to those who already come in with high levels of self-direction and academic confidence.
Getting this right requires deliberate attention to scaffolding, not in the sense of reducing complexity, but in making the expectations, supports, and criteria transparent enough that students from different backgrounds have a genuine opportunity to perform at their best. This is demanding curriculum work, and AI has added another layer to it.
Assurance of learning, capstones, and AI
The arrival of AI in higher education has not created tensions in capstone design; those tensions were already there. What AI has done is expose them more sharply and remove the option of continuing to ignore them.
The assurance-of-learning problem with capstones is longstanding. A capstone may be the primary site where course-level outcomes are assessed, but the conditions in which that assessment happens are often inconsistent, poorly documented, and difficult to moderate. Supervision arrangements vary. Task design varies. Marking criteria are sometimes left at a level of generality that allows wide interpretive latitude. AI invented none of this. But in a pre-AI environment, these weaknesses were less visible, and their consequences were less acute.
AI changes the accountability structure of assessment in ways that are still being worked through. When a student submits a capstone project, the question “Is this their own work?” has always been relevant. Now it is unavoidable, and the answer requires a level of specificity that most assessment frameworks were not built to provide. Distinguishing between supported learning, authorised AI use, and poor AI use requires clear, documented criteria for what each means in the context of a specific task. That clarity is not a luxury; it is now a basic condition of defensible assessment.
More broadly, AI forces a re-examination of what capstones are assessing. If the task is primarily a written artefact that a language model can generate to a substantial extent, then the capstone may be measuring something other than what the course learning outcomes say it measures. This is not an argument against written work; it is an argument for being more precise about which aspects of that work constitute the evidence of learning. Is it the quality of the argument? The methodology? The ability to locate the work within a disciplinary tradition? The capacity to make and justify decisions under conditions of uncertainty? Each of these requires a different task design and assessment approach, and they are not all equally susceptible to AI delegation.
The positive case here is that AI can improve capstone design by forcing the kind of alignment work that should have been done all along. When you must specify what a student is doing in a capstone, not just the output they are producing, but the intellectual and practical work that the output is supposed to evidence, you end up with a clearer, more defensible, and more educationally coherent curriculum object. Tasks designed with this level of intentionality tend to be better tasks by almost any measure.
Building a university-wide model
A decision to develop a university-wide capstone model is a bet that the benefits of structural coherence outweigh the costs of constraint. It reflects a judgment that having a shared framework for what capstones are, what they are supposed to do, and how they should be designed and assessed is more valuable than allowing unlimited disciplinary variation.
That bet is reasonable, provided the model is genuinely flexible enough to accommodate the very different capstone traditions across a university. An engineering capstone looks different from a social work capstone, which looks different from a creative arts capstone. A university-wide model needs to specify what is common across all of these: the structural features, the outcome expectations, and the principles of assessment design, while leaving sufficient disciplinary latitude for each to maintain its integrity.
AI sits at the centre of this work, not as a problem to be managed but as an environmental condition that a well-designed capstone model must address head-on. The questions it forces us to ask about what we are assessing, about what authentic independent performance looks like, and about how we make defensible claims about graduate capability are exactly the right ones to ask as we design capstones fit for the next decade of higher education.
Getting the capstone right has always mattered. Right now, it matters more than ever.
AI Declaration
This post was developed with the assistance of AI. The ideas, research notes, source material, and editorial direction are my own. AI was used to help draft and expand the text from detailed notes and structured headings I prepared. All content was reviewed, edited, and approved by me before publication.
Leave a Reply