As African startups increasingly adopt artificial intelligence to power fintech, logistics, and consumer platforms, many are uncovering a structural challenge beneath the surface. Popular AI agent tools such as AutoGPT and LangChain, originally designed for always online environments, often struggle when deployed in regions with unstable power supply and inconsistent internet connectivity. This challenge, now described as the “Agentic Gap,” is reshaping conversations around how AI systems should be built for emerging markets.
Tobi Adeosun, a software engineer and AI researcher, is among the voices addressing this gap through Contextual Engineering, a new open source framework available via a public GitHub repository, alongside an accompanying book that documents the architecture, patterns, and philosophy behind the approach. His work focuses on redesigning AI agent systems to function reliably in environments where connectivity cannot be assumed, offering practical tools for builders across the Global South.
The problem becomes evident in everyday use cases. In cities like Lagos, even a brief network interruption can cause standard AI agents to fail mid task. These failures often lead to hallucinated responses, incomplete workflows, lost user data, and wasted API credits. For startups operating on thin margins, such disruptions translate directly into increased costs and weakened user trust. Adeosun argues that this is not a user behavior issue but an architectural one, rooted in infrastructure assumptions imported from Western markets.
The Contextual Engineering framework introduces an offline first design philosophy that challenges these assumptions. At its core is a pattern known as sync later, detailed both in the open source codebase and in Adeosun’s book. Instead of sending every user request immediately to the cloud, the system stores actions locally using cryptographically secure queues. These requests are preserved offline and synchronised opportunistically when connectivity becomes available, ensuring continuity of service and zero data loss even in low bandwidth environments.
Another central concept explored in the framework and book is the hybrid inference router. This mechanism allows AI systems to dynamically determine where computation should occur. Simple tasks are handled locally using lightweight on device models, while more complex tasks are routed to advanced cloud models only when necessary. This approach reduces unnecessary API calls, helping startups control costs while maintaining performance and reliability.
Beyond technical implementation, Adeosun frames the work as part of a broader push for architectural sovereignty. He argues that African technology ecosystems must build systems that respect local infrastructure realities rather than waiting for those realities to change. In this view, resilience is not a compromise but a competitive advantage that enables scale, inclusion, and long term sustainability.
Since its release, the Contextual Engineering GitHub repository and book have begun attracting attention from founders, CTOs, and product managers building AI powered products for African and emerging markets. As global discussions around responsible and context aware AI continue to evolve, Adeosun’s work offers a grounded, practical example of how local constraints can inspire better system design.
For Adeosun, closing the Agentic Gap is about more than improving performance metrics. It is about ensuring that the next generation of AI systems works for the environments in which they are deployed. Through open code, documented frameworks, and shared knowledge, his work is helping redefine how intelligent systems are built for Africa and beyond.
