As artificial intelligence becomes increasingly embedded in financial services, identity verification and public administration, debates around governance have largely centred on legislation, regulation and ethics. However, software engineer Abdulqudus Aremu believes the most important governance decisions are made much earlier, during the design and development of AI systems.

Aremu, a software engineer at Seamfix and Chief Technology Officer of AgroHQ, argues that engineers shape how AI behaves long before regulators have an opportunity to intervene.

According to him, governance is often viewed as a legal or policy exercise, whereas many of its most consequential decisions occur during software development.

“We have inherited an idea that governance is something you do to a system from the outside, after it exists. But by the time a system is in the world, most of the decisions that actually govern how it behaves have already been made.”

He said seemingly routine technical decisions, such as setting confidence thresholds in identity verification systems, can have significant real-world consequences.

“Set it too low and you let fraud through. Set it too high and you lock out real people who have every right to access the service. There is no neutral setting.”

According to Aremu, such choices determine who bears the consequences when an AI system is uncertain and should therefore be recognised as governance decisions rather than purely technical configurations.

He also argued that engineers cannot delegate responsibility for the outcomes of systems they design.

“The person who chose the threshold owns the outcome of that threshold, whether they want to or not.”

While acknowledging the importance of regulation, Aremu maintained that legislation alone cannot address design choices already embedded in deployed systems.

“Regulation matters. But regulation is downstream. If the people building these systems do not take governance seriously at the point of design, no amount of policy written afterwards fully fixes it.”

Discussing Africa’s technology landscape, Aremu said the continent has particular reasons to prioritise responsible AI development because digital systems increasingly determine access to identity, financial services and other essential public services.

He argued that errors in AI-driven identity or verification systems can have more severe consequences in environments where citizens may have fewer alternatives or formal appeal mechanisms.

For organisations adopting AI, Aremu recommended practical governance measures rather than complex compliance structures. These include clearly defining which decisions should remain under human supervision, maintaining records of automated decisions, creating accessible appeal mechanisms and testing systems against the needs of vulnerable users rather than average users alone.

According to him, organisations that invest in responsible AI governance are likely to build greater public trust.

“Good governance is not the tax you pay on innovation. In this part of the world, it is a competitive advantage.”

Aremu concluded that as AI capabilities continue to expand, responsible leadership will increasingly be measured not only by the sophistication of technology but by accountability for its real-world impact.

“Building something powerful is, increasingly, the easy part. The harder thing is the willingness to own what that power does once it is in the world.”