The Day the App Stopped Talking — and What I Learned About Listening to Users

A few weeks into developing Arise & Shine Transporters, I encountered a moment that changed how I approach AI collaboration. The AI-powered ops copilot, designed to help dispatchers make decisions based on fleet telemetry and order data, was failing to respond to a dispatcher’s request in a way that made sense. When the dispatcher asked, 'What’s the best route to deliver this order on time?' the AI returned a generic answer that didn’t take into account the specific road conditions or vehicle capacity. It was clear that the AI wasn’t understanding the context of the request.

I turned to Claude to help refine the prompt. I described the situation and asked for a way to phrase the question to elicit a more useful response. Claude suggested breaking the question into smaller parts and providing the AI with more context about the vehicle, the delivery window, and the road conditions. I tried this new approach and saw immediate results — the AI provided a more accurate and actionable suggestion.

The experience taught me that AI, no matter how advanced, still needs precise inputs and clear expectations to function effectively. It also showed me the importance of listening to users and understanding their needs in real-time. The dispatcher’s feedback was crucial in identifying where the AI was falling short.

This insight was applied across the development of Arise & Shine Transporters. I made it a point to involve real users in testing the AI features early and often. Their feedback helped shape how the AI was trained and how it interacted with the system. It was a small but significant shift in my approach to AI development — one that made the product more useful for the people who rely on it every day.

The process was not without its challenges. There were times when the AI still struggled to interpret the nuances of the dispatcher’s language or the complexity of the logistics environment. But with each iteration, the system improved. Today, the AI-powered ops copilot is an invaluable tool for dispatchers, helping them make decisions faster and more accurately than ever before.

This experience reinforced a core truth about building AI-powered products: success depends not only on the technology but also on the people who use it. The AI is just one piece of the puzzle. The real work lies in understanding the user, their needs, and the environment in which they operate. And that, I’ve found, is where the most valuable insights come from.

For anyone building AI-powered tools in Kenya, this lesson is especially important. The environments in which these tools are used are often complex, dynamic, and deeply rooted in local realities. The AI must be trained not just on data but on the lived experiences of the people it serves. That’s what makes the work both challenging and rewarding — and it’s why I continue to build with care, patience, and a deep respect for the people who rely on these tools every day.