Every major education conference in the past two years has had a session on AI. Most of them have followed the same pattern: a demonstration of a tutoring tool, a slide deck showing personalised learning dashboards, and a panel discussion about whether AI will replace teachers.
This is the wrong conversation. Not because these tools are unimportant, but because they address the surface layer of a much deeper structural question.
The Tool Trap
When we introduce AI into education as a product, a tool that sits on top of an existing system, we inherit all of that system's existing misalignments. A personalised learning platform deployed in a school where teachers are not supported to use data, where the curriculum is not connected to assessed outcomes, and where the feedback from classrooms never reaches policymakers, that platform does not transform the system. It becomes another layer of complexity that the system absorbs without changing.
AI does not fix system misalignment. In a misaligned system, AI scales the misalignment.
Where AI Actually Belongs
The more productive question is: where in the architecture of a learning system can AI function as a structural component rather than an add-on? Learning systems need feedback, accurate data flowing from classrooms to teachers to school leaders to policymakers and back. They need adaptation, the ability to adjust delivery based on what the data shows. They need efficiency, the capacity to reach more learners without proportional cost increases. AI can serve all three functions, but only if it is integrated at the governance, data, and delivery layers simultaneously.
The African Context
In Tanzania and across sub-Saharan Africa, the risk is not just that AI fails to improve learning outcomes. The risk is that it introduces new dependencies, new infrastructure costs, and new data governance challenges into systems that are already under-resourced and structurally fragile. Responsible AI integration here means starting with the system's actual feedback and decision-making architecture, and designing AI's role within that architecture, rather than imposing a product and hoping the architecture adapts around it.