Being good with technology used to be relatively easy to describe. A developer could write code. A network engineer could configure infrastructure. Cybersecurity expertise could be demonstrated through knowledge of systems, threats and the tools used to protect against them. None of those abilities has suddenly become less valuable, but AI is making them a less complete description of what technical competence looks like.
Research from CompTIA found that 95% of South African businesses surveyed said digital fluency was becoming either moderately or much more important. CompTIA defines that as the ability to confidently use digital tools, data and emerging technologies, which makes sense until the word “confidently” meets generative AI. These systems are remarkably good at producing answers that look credible even when the reasoning or information underneath them is wrong.
Neil Lund, managing director at Paracon by Adcorp, describes technical proficiency as being able to code, develop systems or configure infrastructure. Technological literacy extends into what happens around that work: interrogating the data informing a decision, recognising a hallucination, spotting bias in an AI-generated output and understanding whether the result actually makes sense in its particular context.
“The ability to construct meaning from data relevant to a business challenge has become as important as the ability to code the solution,” Lund says.
That creates a more complicated hiring problem. Someone can be technically accomplished and still make a poor decision because they trusted bad information, failed to question an automated output or understood the system without properly understanding the problem it was meant to solve.
Technical work increasingly includes judging what machines produce
The World Economic Forum’s Future of Jobs Report 2025 found that analytical thinking remains the most sought-after core skill among employers, with seven out of ten companies regarding it as essential. The fastest-growing skills tell the other half of the story: AI and big data lead the list, followed by networks and cybersecurity, with technological literacy close behind.
There’s no indication here that conventional technical expertise is becoming irrelevant. Software and applications developers remain among the occupations expected to grow fastest through 2030. Instead, the work increasingly combines technical knowledge with the ability to interrogate what technology is doing.
A data scientist can use increasingly powerful tools to surface patterns across enormous datasets, but the software can’t remove the responsibility for deciding whether a pattern is meaningful or misleading. A security analyst can use AI to identify suspicious behaviour more quickly, but still needs enough knowledge of the system to recognise when an automated conclusion is wrong.
Generative AI makes the problem easier to see because the output often arrives already polished. Generated code can compile and still contain a security problem. A summary can read perfectly well while quietly dropping context. An analysis can contain invented information without anything in its tone suggesting that the model has lost its way.
The ability to operate an AI tool consequently tells an employer very little on its own. Chatbots and copilots are deliberately becoming easier to use, while increasingly sophisticated features are being built into software people already know. Learning where to type the instruction may take minutes. Knowing when the answer shouldn’t be trusted can require years of expertise.
South Africa is already struggling to connect skills with jobs
AI is adding another layer to a labour-market problem South Africa hasn’t solved. A World Bank assessment of the country’s digital skills pipeline published in August describes a structural mismatch between high unemployment and digital vacancies that employers struggle to fill. Weaknesses run from foundational digital skills through to the connection between training and the needs of employers, while access to workplace learning remains another problem.
Nearly half of the job postings examined by the World Bank sought transferable capabilities such as communication and problem-solving alongside digital skills. Employers were also struggling to recruit in advanced areas including software development, networking and cybersecurity.
“Soft skills” starts looking like a particularly unhelpful label in this context. Explaining why an AI-generated finding shouldn’t influence a business decision isn’t separate from doing the technical job well. Neither is challenging an assumption hidden inside a dataset or being able to explain a complex problem to somebody outside your discipline.
Lund argues that these capabilities can directly affect the quality of the technical outcome. AI makes that connection harder to ignore because it can move some of the work from producing an answer towards evaluating one.
We’ve already seen a version of this happening outside traditional IT. Generative AI is spreading into South African jobs that wouldn’t previously have been described as technology roles, including creative work and operations. The profession doesn’t necessarily disappear when AI enters it. The skills expected inside the profession begin to change.
The same thing can happen to technical work. A developer remains a developer, but part of the job may increasingly involve reviewing code generated by a machine rather than writing every line manually. That only works if the developer understands the code well enough to recognise when the machine has produced something that shouldn’t make it into a live system.
Employers now have to decide what AI literacy actually means
The language employers use hasn’t necessarily caught up with what they’re trying to hire for. “AI literacy” can be added to a job description quite easily, but the phrase becomes close to meaningless unless the company can explain what competence looks like in that particular role.
CompTIA’s research suggests South African companies know their requirements are moving. Some 81% of businesses surveyed said upskilling and reskilling were a high or moderately high priority, while 59% were using formal skills assessments to identify capability gaps.
Those assessments become much harder when the skill being measured is judgement. It’s relatively straightforward to test whether somebody can write functioning code or configure a particular piece of infrastructure. Testing whether they know when to distrust an apparently convincing AI output requires a task that gives them something plausible enough to question in the first place.
The answer will also differ between jobs. Someone developing software with an AI coding assistant faces different risks from an analyst working with customer information. What employers need to know is how well the person understands the work underneath the automation and whether they can recognise the point at which the technology has stopped being helpful.
Hiring around individual AI products won’t solve that problem either. Models change, interfaces are redesigned and features that currently require specialist tools are steadily turning up inside ordinary workplace software. A candidate’s ability to learn an unfamiliar system and interrogate what it produces is likely to survive those changes better than knowing exactly where every button sits in one current product.
Technical foundations become more important when the software looks clever
There is a tempting conclusion to all of this that doesn’t hold up particularly well: if AI can generate the code or perform parts of the analysis, perhaps deep technical knowledge becomes less important. In practice, outsourcing more of the production to a machine can make expertise particularly valuable when something goes wrong.
A person who doesn’t understand programming is poorly placed to decide whether generated code is secure. Someone without the necessary security knowledge can’t reliably determine whether an automated threat assessment has missed something. The easier these systems become to operate, the easier it also becomes to produce work that exceeds the operator’s ability to evaluate it.
The WEF expects AI and big data, networks and cybersecurity, and technological literacy to be the fastest-growing skills through 2030. Its research also keeps analytical thinking at the top of employers’ existing priorities. Those findings fit together more comfortably than the usual debate about whether technical or “human” skills will win in an AI economy.
South African employers may need people with deeper technical knowledge precisely because AI can now imitate some of the visible signs of technical competence. A polished output isn’t proof that the person using the system understands what happened underneath it.
That leaves companies with a hiring challenge more difficult than adding ChatGPT, Copilot or another AI product to the skills section of a vacancy. They need to find people who understand the technology well enough to use what it produces without surrendering their judgement to it. As generating an answer becomes cheaper and easier, recognising when an answer is wrong becomes a far more valuable part of the job.