| By: Abdulmuizz Olanrewaju
Africa’s technology companies risk producing developers who can ship software faster than ever while understanding less of what they place in production.
A junior developer can now ask a coding agent to inspect an unfamiliar repository, create an interface, connect APIs, build services, write tests and correct errors until the application runs.
Within hours, they can produce work that resembles the output of someone with years of experience, even when they cannot explain the design choices, identify the risks or repair the system without returning to the same machine.
The code may work, but the competence on display may not belong to the person accepting responsibility for it. I call this borrowed competence: access to the output of experience without passing through the intellectual process that makes experience dependable.
It remains hidden while tests pass, then becomes expensive when traffic rises, or an adversarial user finds the assumption that both developer and agent missed.
This matters because today’s junior developers will become tomorrow’s founders, engineering leaders and architects, while their products will shape payments, healthcare, identity, logistics and public services.
If the sector weakens the route through which developers acquire judgment, it will eventually build systems on knowledge nobody inside the company fully owns.
Gwen Davis, writing for GitHub, argues that junior developers should use AI to learn unfamiliar concepts, question generated code and contribute earlier, while warning that AI works better for learning when developers are not under pressure to ship.
I agree that rejecting coding agents would prepare people for a profession that no longer exists. As a full-stack developer working across front-end systems, back-end services, APIs, smart contracts and developer tools, I have seen AI remove repetitive work and shorten the path from an idea to something testable.
Where I disagree is with the belief that earlier contribution naturally produces faster professional growth. Junior developers do not work inside laboratories designed around curiosity; they work inside companies designed around delivery.
AI can raise the standard of a developer’s output without raising the standard of their judgment, allowing the employer to receive better work while the individual gains little ability to solve the next difficult problem independently. That is not accelerated learning. It is performance subsidised by a machine.
Anthropic’s research makes this distinction difficult to dismiss. In a controlled study, participants who used AI while learning an unfamiliar Python library scored 17% lower on a later assessment than those who coded manually, with the largest gap appearing in debugging.
Participants who used AI for explanations learned more than those who delegated the work. AI can extend thought, but it can also offer an escape from it.
Most workplaces reward the escape because completion is easier to measure than comprehension. A developer who accepts the first working answer often appears more productive than someone who spends hours discovering why it is wrong.
Companies tell junior engineers to remain curious while judging them through deadlines and closed tickets, turning “use AI to learn faster” into “use AI so nobody notices what you have not learned.”
Charity Majors argues that software remains an apprenticeship profession and compares generative AI to a fast junior engineer whose work still requires review, context and responsibility.
She is right, but a company can keep junior roles while removing the apprenticeship from them, leaving the employee present while the agent investigates, designs, writes and explains.
My experience with smart contracts makes this risk hard to ignore. A contract can execute exactly as written and still cause irreversible financial loss because the developer failed to anticipate how an adversary might interact with it.
The agent can generate a weak assumption, build the contract around it and produce tests that confirm the narrow behaviour it expected. A developer without independent understanding may see several layers of confirmation where only one original mistake exists.
Simon Willison uses the term cognitive debt for the loss of understanding that occurs when agent-written code grows faster than the human mental model behind it.
Technical debt leaves software that is difficult to change; workforce cognitive debt leaves developers who cannot change the software without asking the machine that created it. Africa’s technology sector needs a new standard for AI-era apprenticeship.
I propose the Explain, Defend and Debug rule. Before AI-assisted code written by a junior developer reaches production, that developer should be able to explain how the system works, defend the major technical choices and debug a failure without handing the problem back to the agent.
Startups should measure whether junior engineers become more independent over time, not only whether they close more tickets.
Bootcamps and universities should test reasoning through live debugging and code defence rather than grading only finished applications. Engineering leaders should protect low-risk tasks as training opportunities, while investors should stop treating smaller engineering teams as proof of stronger companies.
African developers should use AI aggressively, but the sector must refuse to confuse access to code with ownership of knowledge. General-purpose models may not understand unstable connections, delayed payments, shared devices, informal addresses or local fraud patterns as deeply as engineers building for those conditions.
Africa cannot build technical independence on software its developers can produce but cannot fully explain.
If the continent wants more than borrowed competence, it must reward understanding before the debt is collected.
Abdulmuizz Olanrewaju is a full-stack developer with experience across front-end development, back-end services, APIs, smart contracts, blockchain infrastructure and developer tools.
