AI agents are often presented as autonomous, powerful systems that can reason, plan, and act. But what happens when we try to deploy them in environments with unstable infrastructure, uneven data quality, limited compute, and real regulatory uncertainty?

This session strips away the hype around agentic systems and asks a more grounded question: what does responsible agent design look like in Africa?

We will explore practical architectural decisions behind building AI agents that operate within real-world constraints, including:

- Choosing between large and small models for agent workflows
- Designing energy-aware and cost-aware inference pipelines
- Managing hallucination, autonomy boundaries, and human-in-the-loop controls
- Data sovereignty and bias in African contexts
- Governance considerations when agents make decisions that affect people

Rather than focusing on frameworks alone, this talk introduces a systems-level lens for designing AI agents that are sustainable, context-aware, and aligned with local realities.