We Are Filling the Skills Gap With Code Nobody Can Check
AI coding tools help South Africa build more with fewer developers, but the engineers who can catch their flaws are the ones we lack most.

Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
South Africa does not have enough software developers, and AI appears to offer a way out. Coding assistants let smaller teams build more, faster, and companies short of engineers are embracing them. Writing in ITWeb, Eugene Perumal puts the difficulty plainly: the country “is short of engineers who can tell when the model is wrong.” That is the heart of the problem. We are using AI to fill the gap left by missing engineers, yet the code it produces needs exactly the experienced judgement we are short of. The skills gap is quietly becoming a verification gap.
CONTEXT AND BACKGROUND
The numbers behind that gap are stark. Collective X counted about one hundred and eighteen thousand unfilled ICT vacancies in 2025, more than a third of them junior roles. Pnet data shows only forty-two per cent of applicants hold the qualifications most advertised IT jobs require.
AI tools have moved quickly into that gap. Stack Overflow’s 2026 Developer Survey found that sixty-six per cent of developers use AI coding assistants or agents. Yet only six point six per cent would trust AI output for important decisions, and just twenty-four per cent say their employer has published approved tools or formal guidance.
INSIGHT AND ANALYSIS
The danger is that AI code nearly always works while often being unsafe. Veracode’s 2026 testing found that forty-four per cent of AI code-generation tasks produced code with a risky vulnerability, even though almost all of it was syntactically valid. The best model scored worse than last year’s leader. Separate testing by IOActive found nearly a third of generated samples were fully exploitable, and simply instructing a model to write secure code was often ineffective.
The consequences are already visible. The Cloud Security Alliance reports that researchers at Georgia Tech confirmed seventy-four published vulnerabilities linked to AI-generated code by March 2026, with the monthly count rising sharply. It also found that more than seventy-five per cent of developers believed AI code was more secure than human code, while fifty-six per cent said it often introduced security issues. Confidence and evidence are pulling in opposite directions.
I have previously written about this in the context of career development. I argued that when AI does the routine work that once trained juniors, they can produce polished output without understanding it, and the pipeline of future seniors weakens. The security problem makes that concern urgent. The people best placed to catch AI’s mistakes are senior engineers, and we are not producing enough of them. Today’s juniors are tomorrow’s reviewers. If they never learn to read code critically, every year of AI-assisted development leaves fewer people able to judge whether it is safe.
IMPLICATIONS
For boards and technology leaders, the lesson is to stop measuring AI coding by speed alone. Every organisation using AI to write software should set clear rules on approved tools and automated security testing. Human review of AI-generated code must also be an accountable sign-off by a named engineer, not a rubber stamp. Accountability cannot be delegated to the tool. Where code handles personal information, the organisation remains responsible under POPIA, so a leak is not only a technical failure but a legal one.
For educators and employers, the scarce skill is now judgement: reading code critically, spotting flaws and refusing output that does not meet the standard. Xpatweb’s latest survey found forty-seven per cent of employers say AI is already changing the skills existing roles require. Training must catch up, with deliberate time for reviewing and challenging AI work, not only generating it.
CLOSING TAKEAWAY
AI can help South Africa close part of its developer gap, and it would be foolish to ignore that. However, more code is not the same as better software. If we generate faster than we can verify, we are building a security debt that someone will eventually pay, often our customers. The engineers who matter most in the AI era are those with the judgement to say the machine is wrong. Developing them deliberately is how we turn a skills gap into a real advantage.
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