The best AI for African classrooms starts with context, not code

By Oliver Dick, Learning Science Lead at Injini Africa

‍

AI has made it dramatically easier to build something that looks impressive. A founder can move from an idea to a polished prototype faster than ever before. In education, though, that can create a misleading impression: that because we can build faster, we understand the problem better.

We do not.

The first question for an education founder should not be, “What can this model do?” It should be: what does success look like for the learner, teacher or school we are trying to serve, and what is currently standing in the way?

For me, the most useful technology shortens the distance between a person and that success. Everything else is secondary.

‍

Start backwards, not forwards

One of the recurring problems I see in EdTech is insufficient field research. Founders can refine features and plan a route towards subscription before they have properly established what success looks like for the people using the product.

I prefer to work backwards.

If the learner succeeds, what has changed? If the teacher succeeds, what has become easier or more effective? What does the school administrator need to see? Only once those outcomes are clear should we design the journey towards them.

This matters particularly in African education, where a product may encounter different languages, devices, data constraints, classroom structures, curricula and levels of support. These are not problems to “localise” around after the technology has been built. They are part of the product.

That is what “homegrown” should mean.

It is not a protectionist argument that every model or platform must have been invented on this continent. If an imported product is affordable, safe and demonstrably useful, we should use it. Equally, an African founder does not get a free pass simply for being African.

The advantage comes from proximity to the problem: understanding people well enough to notice the things a distant product team may not even know to ask. But proximity only becomes an advantage when founders listen, test their assumptions and change course when the evidence demands it.

‍

Good design often looks simple

One education product I have long admired is ST Math. It is not African. What makes it useful as an example is the discipline of its design.

ST Math teaches mathematical concepts visually and gives learners immediate feedback. It begins with visual ideas before connecting them to symbols and language. The experience also considers more than the learner alone: teachers and administrators need information that helps them support the learning process.

That kind of simplicity is difficult to achieve because it depends on understanding the whole system around the user.

We see this same design discipline in our most promising local ventures across the continent—such as teams deploying offline-first literacy tools or lightweight WhatsApp-based formative assessment loops. They actively resist feature bloat, focusing relentlessly on the core pedagogical bottleneck rather than tech novelty. 

Sometimes founders respond to a constrained environment by trying to solve every constraint themselves. A team building a learning tool starts worrying about becoming a connectivity provider, a hardware company and a content platform at the same time. The result can be a more complicated product that is further away from the original problem.

Before adding new technology, we should ask whether we are using the technology already in people’s hands well enough.

‍

Before you scale, run your concept through this quick checklist:

• The Outcome Test: Can you clearly articulate what changes for the learner or teacher if your product succeeds?
• The Friction Test: Does your AI remove unnecessary administrative or cognitive burden, or does it add another layer of complexity?
• The Constraint Test: Are you designing for the actual device, data, and classroom realities on the ground, or an idealized abstraction?

‍

AI should reduce friction, not add it

This is where AI becomes genuinely interesting for education. Its value is not that every learner should spend more time talking to a chatbot, or that every educational task should suddenly have an AI layer. Its value is in removing unnecessary friction.

AI can help people decode information faster, receive feedback sooner, ask questions without embarrassment and move from information to application more quickly. It can also take over repetitive work that consumes human attention without requiring the best of our humanity.

Education has never only been about access to information. Information is now abundant. The harder task is helping people synthesise it, question it, apply it and make decisions with it.

The same is true for teachers. The goal should not be to automate the teacher out of the classroom. It should be to remove enough administrative and repetitive friction that teachers have more capacity for feedback, judgement, encouragement and connection. Used well, automation can create more room for empathy, not less.

‍

Test the learning, not only the product

A polished prototype and growing user numbers do not prove that an education product is helping anyone learn. Before long-term outcome data is available, founders can still look for meaningful early signals: are learners less frustrated and able to progress? Are teachers gaining information they can act on? Can learners apply their understanding to a new problem?

These signals do not replace evidence of learning impact, safety, affordability and quality. They help a team decide whether it is solving the right problem before it scales the wrong solution.

‍

Build with people, not for an abstraction

At Injini, this thinking sits behind our AI for Education Venture Builder & Incubation Program. Teams move through problem validation, technical design, MVP development and user testing, with learning science and monitoring and evaluation sitting alongside the technical work. That sequence matters. A sophisticated model cannot rescue a poorly defined problem.

The strongest African AI education ventures will not necessarily be the ones with the most advanced technology or polished demos. They will be the ones that understand the people they serve, define success clearly, test assumptions early and use AI only where it genuinely reduces the distance between a user and a better outcome.

Context is not the final layer we add to an education product. In the classroom of tomorrow, artificial intelligence will only ever be as powerful as our understanding of the human being sitting in front of it. 

Start with the context. Let the code follow. 

‍

About the author

Oliver Dick is the Learning Science Lead at Injini Africa, with over a decade of experience in instructional design, digital learning development and educational technology systems. He works with EdTech ventures across Africa on pedagogical auditing, user experience design and evidence-informed product development, with the aim of delivering verifiable learning outcomes. His research interests include applying cognitive science to culturally grounded learning models, evaluating EdTech efficacy in diverse contexts and building scalable pathways for young people to develop future-oriented skills and problem-solving capabilities.

‍