Will AI Replace Graduate Jobs?

The argument about artificial intelligence and employment is usually conducted in aggregates: how many jobs, over what period. The more consequential question is structural, and it is about which jobs.

Entry-level work is a training system

The tasks given to graduates in their first two years are not, for the most part, economically important. They are pedagogically important. Reconciling a report, checking a contract, writing a small well-specified function: these exist as much to teach the shape of the work as to produce output.

That is precisely the category of task that current systems handle well, because it is well-specified, low-context and verifiable.

The individually rational, collectively expensive decision

A firm that stops hiring graduates saves money now and pays later, in a thinner senior pipeline. But it pays later in a market where seniors can be hired from competitors, so the cost is externalised. Every firm faces the same incentive, and the shortage arrives for all of them simultaneously in about five years.

This is a coordination problem, not a technology problem, and it does not resolve itself.

What this looks like in Botswana

Botswana produces a steady supply of qualified entrants into a formal economy that is not expanding to meet them. There were 14,733 graduates in 2024, set out in how many graduates Botswana produces, and the Human Resource Development Council puts the stock of unemployed graduates at 70,512 as at the first quarter of 2024, against 48,076 at the end of 2019.

Treat that stock figure carefully. The Minister of Labour and Home Affairs told Parliament in February 2024 that the number was 40,036. Two official sources, effectively the same period, around thirty thousand people apart, with no published reconciliation between them.

The occupational detail is where this essay’s argument stops being speculative. HRDC lists Information and Communications Technicians among the ten occupations with the most unemployed graduates, rising from 1,557 in the third quarter of 2023 to 3,837 in the first quarter of 2024. For context, and not as a like-for-like comparison, the information and communication industry employed 6,043 people in total — an industry count, not an occupational one, from a different instrument.

The two numbers cannot be subtracted from one another. Put side by side they still say something a prospective entrant needs to hear: the pipeline into this field is filling faster than the field is absorbing it, before any of the automation effects discussed here have been attributed. We set the sector’s actual size against its promotional framing in how big Botswana’s ICT sector actually is.

Why the spending does not settle it

It is not a question of under-investment in education. The 2026 Budget Speech records that Botswana allocates an average of 7.1% of GDP to the sector, "among the highest levels for upper middle-income countries in the region", and in the same paragraph states that "education outcomes have not kept pace with this level of investment, resulting in a persistent mismatch between spending and performance".

That is the government’s own diagnosis, and it describes a system that can keep producing credentialled entrants while the thing that converts them into practitioners — the first two years of supervised work — gets scarcer.

What would actually help

Redefining junior roles around verification and judgement rather than production
Apprenticeship structures where the training cost is shared rather than borne by whichever firm hires
Employers being explicit about what evidence they accept from candidates, so preparation is not guesswork

None of these is speculative. All of them require someone to act against a short-term incentive, which is why the default outcome is that nothing happens until the shortage is visible. The measured evidence on what AI does to entry-level work is set out in what junior now means, what most firms are actually doing with AI, and why the 40% African outsourcing figure is wrong.