Moving From Skills to Capabilities

• If students outsource critical thinking tasks to AI, they can succumb to metacognitive laziness and fail to experience genuine learning.
• Graduate students tend to place a high value on gaining capabilities, such as mastering the tools and templates that help them translate frameworks into practice.
• Schools should encourage students to engage in self-regulated learning, in which they assess their work and hone their judgment—without the aid of AI.

 
Business schools constantly adapt curricula to meet labor demands for skills. But are business programs helping students develop the capabilities employers want?

Skills are explicit procedures that students can learn in the classroom. Capabilities are abilities that individuals must develop so they are able to act when no set procedure applies. Students can practice building these abilities in the classroom by making decisions and confronting consequences.

It’s crucial for students to take these steps because the work world is changing. The 2025 Future of Jobs Report from the World Economic Forum predicts that 40 percent of job skills will shift by 2030 and artificial intelligence will affect 86 percent of businesses. The most in-demand capabilities will be those that AI cannot perform: making judgments under ambiguity, engaging in continuous learning, making sound decisions in new contexts, and doing the right thing when no procedure exists.

The Performance-Versus-Capability Gap

One of the most essential capabilities in today’s workplace is critical thinking—but there’s a risk that students are starting to outsource it to AI. Until recently, graduate management programs have depended on students to integrate theory and practice through their own efforts. But today, AI can perform this integration for students almost instantly. This results in improved surface performance, but leaves no evidence of genuine learning.

This growing gap between performance and capability is now empirically observable. In a randomized experimental study conducted in 2024, Yizhou Fan and colleagues compared 117 students who participated in an English reading and writing task. Students were divided into four groups: those who had support from ChatGPT, those who had support from human experts, those who were guided by a writing analytics checklist, and those who had no support.

Students in AI-supported groups saw the most improvement in essay scores—even better than students who had the help of professional tutors. Fan and his co-authors noted that this uptick “might be the result of ‘AI-empowered learning skills,’ which improve performance but at the expense of developing real human skills.” In fact, evidence from the study showed that students in the AI group showed no knowledge gain or transfer. The performance of these students improved, but their capability did not.

The most in-demand capabilities will be those that AI cannot perform, such as making judgments under ambiguity and doing the right thing when no procedure exists.

Process mining explained why. Students in the human-expert group moved slowly to orient themselves to the task, evaluate their progress, and reread the source before revising their essays. This allowed them to experience genuine learning. By contrast, those in the AI group fell into a tight loop of consulting the model and revising, so their work improved but their understanding did not. The researchers use the term metacognitive laziness to describe this pattern of relying on AI to do the work.

Metacognitive laziness is not unique to one region. 2025 research by Yuan Yao and colleagues found the same pattern among Hong Kong postgraduates who were using ChatGPT. In Pakistan, Mashaal Sabqat et. al. studied 391 medical and dental students for a 2025 report that found a Spearman correlation of 0.621 between AI reliance and metacognitive laziness. Of these students, 62.4 percent worried that AI could harm their future patient care.

Students preparing for high-stakes practice already recognize the gap between performance and capability in their own learning. In a 2026 study of 422 Turkish university students, Eyüp Yurt and Ismail Kuşci identified the mechanism: AI might not have a directly negative effect on critical thinking, but it encourages epistemic laziness and metacognitive weakness. The problem occurs when students stop doing their own work, especially if programs don’t teach them how important it is that they keep doing it.

Focused on the Wrong Solution

To address this core problem, the leaders of graduate business programs need to rethink how they integrate AI into their curricula. Currently, most graduate programs use AI to deliver courses and provide feedback. But if students are going to retain the ability to think independently, schools cannot simply provide more access to AI. They need a different demand-side response.

In a 2025 article, Xiaoqing Xu and co-authors found that students who have explicit metacognitive support alongside AI make measurable gains. Problems arise when their use of AI lacks metacognitive scaffolding. Most graduate programs have not built such scaffolding, and even where they have, the work of operating it falls to the student, not the program. A program can teach the practices, but it cannot do the metacognitive work on the learner’s behalf.

Self-regulated learning requires students to assess their own work against specified standards—and to take these steps before and after consulting AI.

In a 2023 article, Jason Lodge, Paula de Barba, and Jaclyn Broadbent made a similar point. They noted that AI literacy and critical thinking, as usually defined, are not enough in the face of generative AI. The key capability that sets a learner apart from a machine is self-regulated learning. This is an acquired ability in which judgment is essential. It requires students to assess their own work, predictions, and reasoning against specified standards—and if they’re using AI, to take these steps before and after consulting the tool.

Self-regulated learning should not just be a part of the curriculum. It should be a personal discipline that students maintain throughout the program, with or without faculty input. Programs can introduce and support the concept of self-regulated learning, but they cannot do it for students. That is the point.

What Students Identify as the Gap

With the goal of helping working graduate students develop self-regulated learning habits, I have created a five-practice discipline that enables students to build capabilities without over-relying on AI. I describe this discipline in an article I wrote for University World News. This is just one approach; schools could devise other ones that work just as well.

Altogether, the practices I have devised take students about two to three hours a week. The core is a 30-minute Friday session in which students use their own words to answer four questions: What happened this week? What assumption proved wrong? What principle does this show? Where would this principle mislead me?

It is my belief that students have long been aware of the same gap these practices address. Between 2018 and 2025, I taught cohorts of master’s students at CFVG (Centre Franco-Vietnamien de Formation à la Gestion) in Ho Chi Minh City. Whenever I asked them about the value of their courses, the same answer kept appearing: They valued learning how to translate frameworks into practice.

One student looked for “specific tools and templates that help me implement the general knowledge from the book to the practical work.” Another appreciated the way a course in the EMBA program “brought a great balance between theory and real-world practice.”

Students asked for more practical anchoring, such as case studies and workplace examples, so they could apply frameworks in practice.

When I inquired what more they wanted in their classes, they did not ask for more theory. They asked for more practical anchoring, such as case studies and workplace examples. These professionals already had frameworks. They wanted experience applying these frameworks in practice as part of the program, not after class when they had no support.

Their comments make it clear how schools should design their programs in 2026—in ways that help students build capability, whether or not they’re using AI.

What This Means for the Capstone

There’s a clear implication for how schools supervise the thesis portion of capstone courses. Over the course of the year, students produce weekly syntheses, dated data decisions, and revisions of their personal operating manuals. Almost as a side effect, these outputs create primary-source data sets that are grounded in first-person observation. Thesis supervisors can use these outputs to determine if students are demonstrating capability rather than simply reporting the findings they’ve gathered from surveys.

Supervisors also can turn to methodologies such as action research, reflective practice, and autoethnography to gauge if students are developing capabilities. One recent example was described in an article in the International Journal of Qualitative Methods, which looked at how master’s theses by students in a Colombian innovation program were driven by action research. Students demonstrated capability by using dated evidence to show what they could do at the end of the program that they were unable to do at the program’s start.

It’s likely that employers and peer institutions will value such proof of competence more than they would value the type of literature review that is a standard feature of master’s theses.

The Risk Ahead

While my personal observations are drawn from one institution, where I teach to mostly Vietnamese students, additional findings from China, Hong Kong, Pakistan, and Turkey support my conclusions. The recurring patterns demonstrated in these countries make it urgent for graduate management program administrators to question how they integrate AI into their curricula.

Such a question might be uncomfortable for schools that allow students to depend heavily on AI tools. However, program leaders must be aware that, according to research, AI improves performance but harms the cognitive processes that performance is supposed to measure. Therefore, school leaders must design their programs to provide graduates with the capabilities they will need, not just the skills.

Institutions face a real risk when they keep emphasizing the skills the labor market increasingly signals it does not need while they fail to build the capabilities that employers want.