Across every lecture hall and technical & vocational education & training (TVET) workshop from Johannesburg to George, invisible assistants — ChatGPT, Claude, Copilot — are redrafting essays, solving equations and producing first-cut financial models in real time, in students’ pockets, during class.
Artificial intelligence (AI) has been sitting in the front row for two years. The only question left is whether the institution teaches alongside it, or is caught drafting the policy to ban it.
Institutions that frame this as a cheating problem have already lost the argument. The consequential question is different: in a world where knowledge is freely available to anyone with a data bundle, what is the value proposition of a qualification?
The strategic mirror
Our universities and TVET colleges hold genuine strengths: physical infrastructure, national accreditation, social trust and unmatched scale.
The weaknesses are equally structural:
- curriculum cycles obsolete before a student reaches second semester;
- assessments that test recall over reasoning; and
- a work-integrated learning system that exists on paper more than in practice.
The opportunity is capacity reclamation. Educators using AI at least weekly save 5.9 hours per week — six full working weeks annually, according to Gallup’s 2025 national survey. Schools with a formal AI policy see 26% bigger time savings than those without one. The threat is quieter.
Mercer’s 2026 Global Talent Trends report found 53% of South African employees fear their skills will become obsolete. TVET graduates who can recite tax law but have never reconciled a live ledger in an AI-integrated system do not face a skills gap. They face an employability ceiling.
Policy as the new pedagogy
The policy conversation in most institutions begins in the wrong place — detection. How do we identify AI-generated work? That is a rear-guard action against an irreversible tide. The right question is what is assessed and why.
Testing a student’s ability to recall information in 2026 is the academic equivalent of testing a pilot on their ability to flap their arms. Assessment by application is the governing principle that replaces it.
Testing a student’s ability to recall information in 2026 is the academic equivalent of testing a pilot on their ability to flap their arms. Assessment by application is the governing principle that replaces it.
When AI use is permitted — and it is, because it is the tool of the trade — the grade reflects what the student adds on top of the output: the quality of the prompt, the critical review of the first draft, the analytical leap taken thereafter.
The institution is no longer reproducing encyclopedias. It is producing curators and orchestrators of intelligence. That distinction belongs in every programme qualification mix.
From literacy to fluency
The real bottleneck is not student readiness. It is the front of the classroom. AI literacy means knowing a chatbot exists. AI fluency means deploying one to solve a problem no-one has defined yet.
A 2025 financial sector survey found 53% of staff feel insufficiently trained to use AI — a gap that almost exactly mirrors tertiary faculties. Fluent lecturers do not fear the technology. They weaponise it: generative AI handles rubric creation, slide preparation and assessment drafting in a fraction of the time, returning hours that belong to the most important design decision in contemporary education.
Crucially, fluency also means orchestrating multiple tools, validating outputs and embedding real-world constraints into learning design. The answer to what to do with that time is unambiguous: stop lecturing and start simulating.
Creating the industry-ready human
An accounting student who has only read about a loan application is a liability. Give them a messy data set from a distressed SME, have them use AI to run the credit risk analysis, and put them in front of actual financiers to defend their investment memo. That is the baseline, not an enrichment exercise.
A mechanical engineering student who uses AI-driven computer-aided design tools to optimise a 3D-printed part, then troubleshoots the physical failure in the workshop, arrives at their first job with something no lecture can manufacture: digital muscle memory.
AI performs at its ceiling only when the data feeding it is clean, current and contextually relevant — garbage in, garbage out remains the most expensive lesson in enterprise technology
This is the “doing economy”. Industry pays for what graduates can do with what they know, under pressure, with the tools already on the desk. Integrating the actual AI systems used in the sector — diagnostics in automotive workshops, predictive analytics in retail management, AI-assisted coding environments in software houses — is not a curriculum enhancement. It is the minimum viable qualification.
The architecture behind effective AI adoption is unglamorous but non-negotiable. AI performs at its ceiling only when the data feeding it is clean, current and contextually relevant — garbage in, garbage out remains the most expensive lesson in enterprise technology.
Learning management systems must be the integration spine: attendance, assessment performance, at-risk flags and micro-credential completion feeding a single student record that informs the facilitator and the funder.
Layered onto that is cybersecurity hygiene — student data is personally identifiable, Protection of Personal Information Act-regulated, and a live target. None of this is deployable without internal human capacity that understands the tools and the obligations.
The sovereign student
Filling a vessel with static data is no longer an educational outcome. It is a liability. The graduate who wins in the next decade will not be the one who carries the most knowledge — it will be the one with the cognitive agility to learn, unlearn and relearn at the speed the market demands.
Stellenbosch University’s AI² approach — shifting focus from policing AI use to fostering transparency in AI-assisted work — is already producing that graduate. The blueprint exists. The tools exist. The demand is documented and urgent. All that remains, quite frankly, is institutional will.
• Mafinyani is risk advisory & financial modelling partner at DiSeFu, a specialised financial technology and risk advisory firm operating in the sub-Saharan region.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.