We cannot find AI specialists because we do not keep the ones we produce
AI skills have entered the country's hardest-to-fill list while our universities quietly hand-finish talent to the world.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
A young engineering graduate at the University of KwaZulu-Natal has built a system that reads South African Sign Language through a camera and speaks it aloud. Akhil Hansrajh developed it as his final-year design project, training a computer vision model on a dataset he had to assemble himself because none existed for the local dialect, and his supervisor described it as among the first assistive technology projects of its kind in the discipline. Hansrajh grew up with two deaf parents and watched them struggle at bank counters, government offices and social grant queues. The story has been reported as a moment of national pride, and it deserves to be. The harder question sits just underneath it. We have produced a capable young engineer who solved a real South African problem, and we have almost no mechanism to keep him, fund him, or deploy what he built.
CONTEXT AND BACKGROUND
For the first time in nearly a decade, artificial intelligence specialists have entered South Africa’s top ten hardest-to-fill skills. The 2026 Critical Skills Survey found that 22% of participating employers reported difficulty recruiting AI and emerging technology specialists, placing the category fourth among the most sought-after occupations, with these roles commanding a global wage premium of up to 62%. The standard reading of that figure is that our universities do not produce enough people.
The window in which we could keep someone like Hansrajh is narrower than that reading admits. A study of 437 IT professionals who approached one immigration firm about relocating to Australia between April and June this year found that roughly 70% wanted to leave immediately or within six months, and nearly half were aged between 25 and 34. That is the exact cohort in which specialist expertise forms, and the exact cohort a newly graduated engineer is about to join.
INSIGHT AND ANALYSIS
The pipeline does not break in the lecture hall. It breaks at graduation, when a highly commended project is filed, the supervisor moves on to the next cohort, and the graduate enters a market that pays far more for routine enterprise work than for the thing he has just proved he can do.
That gap has a name in policy circles. South Africa spends roughly R30 billion a year on research and development, and the head of the Technology Innovation Agency has said plainly that much of it disappears into what is called the valley of death, the stretch between a working prototype and a product anybody can buy.
The demand side is not theoretical either. South African Sign Language became the country’s twelfth official language in 2023. Every department serving the public now carries an obligation it has no practical means of meeting, and a student has demonstrated part of the answer without a single procurement conversation taking place.
IMPLICATIONS
There is movement. The Technology Innovation Agency has restructured towards commercialisation rather than project funding, committing R473 million to venture capital funds and roughly R62 million to sovereign AI work. Whether any of that reaches a graduate with a working prototype depends on the test applied at the door. Conventional venture capital and most grant criteria ask for revenue, traction and a registered company, none of which a final-year project can show. What closes this particular gap is different instrumentation: pre-commercial procurement, where a department commits to buy a first imperfect version, and seed-stage bridge funding routed through university incubators that can carry a prototype for the eighteen months before it is investable.
Business leaders should stop treating the skills shortage as a supply problem to be solved by importing specialists. The faster intervention is absorption. Fund final-year projects that address local problems, and hire the graduate who built something over the one who lists the right frameworks.
Universities carry the other half. A project that earns an award and then stops is an institutional failure, not a student achievement. Technology transfer offices need to treat assistive and local-language work as commercially serious rather than as goodwill.
CLOSING TAKEAWAY
We keep describing our AI problem as scarcity. It is closer to leakage. The talent forms, proves itself on a problem nobody else would touch, and then meets a country with no mechanism to catch it. Akhil Hansrajh built something that could serve a community our institutions have failed for decades. What happens to him over the next two years will tell us more about South African AI capability than any policy document. The measure of a working pipeline is not how many graduates we produce. It is how many of them are still here, building for us, in five years.
Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments