Africa Skills Week is in Harare this week with a roadmap for turning training into industrial value. The plants that would do the teaching are still being negotiated, and the skill that matters has never been transferable by certificate.
The African Union is holding Africa Skills Week in Harare from 12 to 16 October, under the theme of skills for resilient and competitive African industries, from training to value creation. The programme runs to more than 600 international delegates, over a thousand local ones and sixty exhibitors, with a plenary on African Continental Free Trade Area value chains on the Thursday and a communiqué adopted on the Friday.
The diagnosis behind it is correct, and it is one this series has made repeatedly. Green industrial strategies in Africa are written around factories, finance and minerals, and they are short of the technicians, maintenance specialists, power electronics engineers and process controllers who make any of it run.
The difficulty is in the direction of travel the week assumes. Training to value creation reads as a sequence: build the training system, and industrial capability follows. The economics of how industrial skill actually accumulates says the arrow mostly points the other way.
Arrow's 1962 paper on learning by doing began from an awkward empirical regularity in airframe manufacturing. The labour hours required to build a given aircraft fell steadily and predictably, not with calendar time, not with training expenditure, but with the cumulative number of units already built. Experience was a function of output. Productivity improvement was a by-product of production itself.
Arrow's point was that this is not a detail of aviation. It is a general feature of how technical knowledge in production is created: much of it is not written down anywhere, cannot be taught in advance of the activity, and is generated by doing the thing repeatedly and encountering the problems the thing throws up. New knowledge, in his framing, comes from attempting to solve problems, and the supply of problems comes from operating.
Apply that to green industrial capability and the awkwardness is immediate. The skills that the African Union's own agenda names, battery-system maintenance, power electronics, industrial automation, quality assurance, process engineering, are almost perfectly the set that Arrow's mechanism governs. None of them is a body of facts. All of them are pattern recognition built on a long tail of specific failures: this inverter trips under this load at this ambient temperature, this cell chemistry drifts in this way, this line stops for this reason on the Monday after a long weekend.
A classroom can supply the theory. The fault is what does the teaching, and faults are produced by plants that are running.
Tanzania is a reasonable test of the ordering, because it is currently assembling exactly the kind of industrial base the skills agenda is meant to serve. The government is advancing two critical-mineral agreements, continuing to negotiate a $42 billion LNG development, and targeting around 8,000 MW of generation capacity by 2030, a very large increase on today.
Those are the assets that would generate the learning. They are also, every one of them, in negotiation, financing or construction. A cohort trained in 2026 against a plant that commissions in 2030 spends four years not accumulating the only thing that makes them valuable, and the half-life of unpractised technical skill is considerably shorter than four years.
This is where training systems and industrial policy quietly fail each other. Each is on its own clock. A skills roadmap can be adopted in a communiqué on a Friday. A smelter takes six years. The roadmap will be measured on graduates by then, and the graduates will be measured on nothing, because there was nothing for them to be measured against.
If learning is a function of cumulative output, then graduate counts are the wrong object entirely. They measure throughput of a training system, not accumulation of capability, and the two diverge most exactly where the industrial base is thinnest, which is where the programmes are aimed.
Three quantities would actually track it.
Hours on operating equipment per trainee, not hours of instruction. Equipment that is running, in a commercial setting, with consequences when it stops. A workshop rig that nobody depends on produces familiarity, not judgement.
The in-house share of maintenance over time. When a new plant opens, maintenance is done by the vendor or by flown-in specialists. The question is what that share looks like in year three and year six. A falling share of imported expertise is capability accumulating; a flat one means the learning is happening inside the vendor's organisation rather than the country's.
Retention in the role, not placement into it. Placement is a transaction. Capability is cumulative output, so a technician who leaves after eight months has accumulated almost nothing and taken it with them. Placement rates can rise while capability does not move at all.
The briefing framing for this week proposes measuring the proportion of graduates entering relevant technical employment and whether local firms become less dependent on imported expertise. The first of those is the weaker indicator and the one that will be reported. The second is the real one.
This series has already argued that technological capability accumulates slowly and firm by firm, following Lall. Arrow is a different claim and the policy consequence diverges.
Lall's capability is built deliberately. A firm invests in engineering effort, absorbs, adapts and gets better because it set out to. The implication is that capability-building is an investment decision and can be supported as one. Arrow's learning is largely involuntary. It happens because production happens, whether or not anyone intended it, and it cannot be bought separately from the output that generates it.
Both are true and they point at different instruments. Lall says fund the deliberate effort. Arrow says the output has to start, and until it does there is nothing for the deliberate effort to work on. A skills programme running four years ahead of the plant is being asked to substitute for the second while being funded as though it were the first.
Follow one cohort past placement. Three years out from a technical qualification in a country with a thin industrial base, asking what they are doing now and how much of it uses what they trained on. This is not expensive and as far as we can establish it is not done.
Compare two plants of similar vintage. One with a vendor maintenance contract, one that staffed maintenance internally from the start, and track the in-house share and the unplanned downtime in both. That is a direct test of where the learning is landing.
Ask the maintenance crew what they wish they had known in month one. The gap between that answer and the curriculum is the content a training system can realistically supply, and our interview guide sets out how to collect it without leading.
For governments, the conclusion is that the skills programme and the plant need to be on the same clock, and that a programme which cannot name the specific operating assets its graduates will work on is producing certificates. For DFIs, it is that the maintenance function in the first three years of operation is a capability instrument that nobody is funding as one. For investors, it is that the in-house maintenance share is a better read on a country's industrial trajectory than its enrolment figures, and it is already being measured inside every plant for other reasons.
If this reads close to a decision you are weighing, we would like to hear about it. A short note is enough to start.
The Lab is an independent research team, not an advocacy organisation. We do not campaign, endorse products, or accept commissions where a finding is expected in advance. Where a partner's own theory of change is what needs testing, that is a legitimate engagement, and the Lab tests it honestly.