Most conversations about AI and the economy start the same way: how many jobs will be lost? It is an understandable question. But it is the wrong starting point if you want to do something useful. A net jobs figure reveals almost nothing about the underlying reality or what actions to take. Two cities might have identical employment stats yet be experiencing very different underlying changes. One might be developing new skills and creating innovative jobs, while the other could be quietly losing ground.
Generative AI is a general-purpose technology: it refers to artificial intelligence systems that produce new outputs. Like electricity and the internet before it, it lowers the potential cost of a broad class of tasks across the whole economy. What makes a general-purpose technology unusual is that each improvement unlocks applications across sectors that had no reason to engage with it before. The gains spread like ripples. Sooner or later, they reach you.
Part of the difficulty is that the impacts of AI aren’t always as visible as AI itself. A South African firm may never use a single AI tool and still find itself undercut by competitors whose products and processes were quietly shaped by it. Watching only for AI adoption misses most of what matters. The risk is not a wave of layoffs directly from AI. It is the loss of competitive advantage from not using what Generative AI offers firms, industries and value chains. To understand what’s really happening, you need multiple perspectives.
Five observation spaces, not five steps
The Technological Change Observatory works with five observation spaces, which are interconnected rather than arranged in order. A change in any one of them can influence what’s possible in all the others. When a company begins using AI, it naturally shifts the expectations it sets for its suppliers, employees, and supporting institutions. Conversely, an institution that doesn’t keep pace can limit a company’s options, no matter how eager the company is to adapt.
The micro space is where transactions and decisions happen
In this observation space, we observe how firms are reorganising their work, the workers within those firms, and their relations with suppliers, customers, and service providers. Shadow AI use without official permission is one of the most important signals. It is also extremely hard to measure because it is largely hidden. Shadow AI use tells you where people see value, where official systems are falling short, and where organisational policy is running behind reality. But we can also track how companies formally disclose their Generative AI use. A firm that buys an AI subscription and changes none of its workflows has not absorbed AI; it has bought a subscription.
The industry and value chain space is where change propagates
In this space, we observe how industries reorganise because of technological adoption. No firm operates alone. When a lead bank automates underwriting or a mining major adopts predictive maintenance, the downstream effects on BPO providers, engineering services clusters, and equipment suppliers often exceed the direct impact on the adopting firm. Incumbents typically miss new competitors not because they are inattentive, but because the threat arises outside their existing frame of reference. Industries that focus only on their known competitors are already at risk from new competitors entering from unexpected markets.
The meso space is where generic policy becomes selective support
This is the layer where technological knowledge is identified, translated into practice, and disseminated. Without it functioning well, individual firms cannot build the capacity to absorb new technology, and the gains from a general-purpose technology go elsewhere. Examples of meso institutions are standards bodies, technology extension services, industry associations, and universities: these organisations identify persistent patterns that neither the market nor broad policy can address alone. They carry signals upward to policy and sideways to other organisations that support innovation, sector development, or provide public goods. The problem in South Africa is not that these institutions are absent. It is that they are not connected in ways that allow what they observe to travel between organisations.
Shared infrastructure is also a meso function, as it reduces the costs of doing business. Very few South African firms can build AI models suited to local languages and market conditions on their own. India understood this early. Its 2025 the India AIKosh initiative provided a national repository of AI datasets and models across 20 sectors, with high-performance computing capacity at roughly one-third of global hyperscaler rates through its IndiaAI Mission pool. South Africa has made similar commitments at the G20 level. However, as far as I can tell, the infrastructure does not yet exist.
The governance and social mediation space is where structural change gets negotiated
This space is where different stakeholders plan and negotiate shared priorities. Examples include labour legislation, sector development plans, and social compacts between government, business, and labour. These fora play an important role in preparing multiple stakeholders for technological disruptions and change. Or they can create defensive formations that resist change. The question is whether these platforms position South Africa to move quickly in new opportunity spaces. For instance, can these dialogue platforms explore both how to rapidly improve productivity and how to protect vulnerable jobs? Generative AI will require simultaneous conversations.
The geopolitical and geoeconomic space is where external constraints become visible
This observation space focuses on South Africa and its relationship to the global economic and technological landscape. South Africa does not control where AI compute infrastructure sits, who owns the foundation models, or how technology policy shifts in Washington or Beijing affect our options. Generative AI is being embedded into the smartphones, sensors, machines, and software platforms we import, arriving whether our policy and strategies are ready or not. That is not an argument for waiting to see what will happen. It is a constraint that an honest industrial policy dialogue has to name.
Observation spaces are not outcome indicators
Outcome measures, employment statistics, investment flows, and market concentration ratios tell you what has already happened. By the time a worsening labour market shows up in official statistics, the causes are often two or three years old.
Observation spaces give you an earlier warning. Most current AI policy monitoring focuses entirely on the outcome column while leaving the observation spaces unwatched.
Why this moment is different
Generative AI has not yet settled into a stable dominant design. Multiple technical approaches are still competing. The window for influencing South Africa’s position is open and it will not stay open indefinitely.
Countries further along are not waiting for better data or more certainty. The hardest gap to bridge is between government departments and the industries they are meant to support. Each is watching a different part of the system, with different incentives and different timeframes. Without something operating in the space between them, the signals that each holds never combine into a picture that either can act on.
South Africa has the institutions. What is missing is the connective tissue between them.
Who does the observing?
It cannot be one organisation. Technological intelligence needs to be distributed across firms, industry associations, research organisations, technology extension centres, and government programmes. The more people actively looking for signals of change, the better.
This is the idea behind the Technological Change and Innovation System Observatory at TIPS, which is co-implemented by Mesopartner. If you are part of a government programme, an industry association, or a firm that wants to build this kind of AI observation capacity in your sector, the tools exist. Get in touch to find out how we can work together.
This post was published as part of my work at the Technological Change and Innovation System Observatory hosted by TIPS, and co-implemented by Mesopartner.
Image Credit: Cape Town, Western Cape, South Africa. Photo by K from Pexels https://www.pexels.com/photo/aerial-view-of-cape-town-skyline-at-twilight-33622091/