Adewole Ampitan Adeyemi is an experienced digital product Designer with over seven years of expertise delivering fintech, AI and scalable digital products. Throughout his career, he has designed high-impact products for leading companies including Interswitch Group, Cyberspace, DOT, and Descinder. Driven by a passion for community and mentorship, Adewole contributes to the global design community as the founder of DesignFlow, an open-source project that guides UI/UX designers through the end-to-end product design process. In this interview with OPEYEMI BABALOLA, he argues that the future of Artificial Intelligence in product design will depend less on speed and automation but more on how well products understand the realities of the people using them.

How has real user behaviour challenged what AI systems assume about users?
One important thing we must understand in the rise of AI systems within product design is that real user behaviour cannot be abstracted away or treated as secondary. That is where many assumptions begin to fail and I would share some personal experience with you. While working at Interswitch Group on Quickteller Business and agency banking platforms, especially during usability sessions and field observations, I repeatedly encountered behaviours that the data alone could not fully explain. For example, fraud detection models often flagged shared devices as suspicious activity. But in reality, one phone might serve an entire household, or belong to an agent processing transaction for dozens of customers in a single day. Similarly, some credit and behavioural models assumed users received structured monthly salary deposits, whereas many users actually earned income daily, mostly through cash-based transactions. Those experiences revealed a clear gap between model assumptions and lived reality. Many AI systems are designed around the assumption that users are individual, continuous, and consistently measurable. But real users are often collective, intermittent, adaptive, and only partially observable. Their behaviour is shaped by context, infrastructure, culture, and economic realities that structured datasets do not always capture well. That is why UX research and direct observation remain essential, even in AI-driven systems. The discipline of getting close to real behaviour is often what reveals the difference between what the model interprets as a signal and what is actually happening in people’s lives.

How does designing for African users differ from designing on global data patterns?
This question is close to my heart, because most of my UX career has been spent on exactly that gap. Designing for African users, especially from direct ethnographic research in Lagos markets, taught me that behaviour is deeply shaped by environment and context in ways global data patterns often fail to capture. Global product assumptions usually come from highly connected environments where infrastructure is stable, device ownership is personal, and digital trust is already established. But in many Lagos markets, I observed users operating under very different realities. For example, I met merchants sharing devices, relying on multiple SIM cards, checking transactions with neighbours before trusting confirmations, and developing personal workarounds for unstable connectivity. These behaviours would look unusual or inefficient if you only viewed them through analytics dashboards. But in context, they made complete sense. One major difference is trust. Many global systems assume that once a transaction says “successful,” the user moves on. But in our market observations, trust was social and behavioural. Users often needed reassurance beyond the interface itself. I also learned that African users are incredibly adaptive. They do not necessarily use products the way designers expect. They improvise constantly around infrastructure limitations, cost, connectivity, and familiarity. That experience changed how I approached design completely. I learned to design mobile-first not as a trend, but as a constraint that sharpened every decision. A 2 MB onboarding video is not elegant; it is a refusal to serve the user. Designing for sub-3G environments, low-end devices, and shared phones forces clarity. It strips products down to what actually matters. Ironically, those constraints often produce better products for everyone, not just the original audience. Designing purely from global patterns can create products that scale technically while excluding people socially. What African users expose are not edge cases, but weak assumptions hidden inside many modern products. Their realities force designers to confront questions about trust, resilience, accessibility, and adaptability that apply far beyond Africa itself. That was the biggest shift for me. I stopped seeing African constraints as limitations to design around and started seeing them as pressures that reveal what good design actually requires.

What gap was DesignFlow Kit solving, and how is your approach different?
I developed DesignFlow Kit as an AI connector rather than simply an AI generator. Instead of just producing screens or artefacts, it connects research insights, design decisions, and delivery workflows so that when one-part changes, the others remain aligned. That distinction is what makes it different. After nearly eight years leading design across Cyberspace, DOT, and Interswitch: running sprints, design reviews, handovers, and governance processes, I kept seeing the same pattern repeatedly. As design systems and AI tools became better at generating outputs, the real bottleneck was no longer production. The harder problem was maintaining coherence as products evolved. For example, Research findings would get lost between sprints. A user insight uncovered in week two would quietly disappear by week six. Design decisions made during workshops would never fully make it into engineering handovers. Design system rules drifted out of sync faster than teams could maintain them. The issue was never the absence of artefacts. It was the breakdown of continuity between them. DesignFlow Kit was built to address that gap.

It creates traceability between research, design rationale, system standards, and delivery tasks so teams can move quickly without losing context along the way. I also made a deliberate decision to release it as open source. I wanted teams in Lagos, regional NHS Trusts, or anywhere else in the world to be able to adapt it to their own operational realities, governance structures, and data constraints.

Many closed AI design tools quietly encode Silicon Valley assumptions about users, infrastructure, trust, and behaviour. Those assumptions do not always translate well across different environments, especially in places like African markets where the realities of connectivity, shared devices, and trust operate differently. Existing AI tools mostly help individual designers produce screens faster. DesignFlow Kit approaches a different problem entirely: helping entire design teams stay aligned while moving quickly.

Most significant contribution in relation to scale or user impact?
The contributions I’m most proud of sit in two areas. The first is my work at Interswitch, where I co-led ideation and feature design for Quickteller Business and the agency banking platforms, products that now serve millions of users across Nigeria. What made that work significant was not just the scale, but the kind of users it reached. The USSD, QR payment, and withdrawal flows I designed were often being used by underbanked individuals interacting with digital financial services for the first time. At that scale, improving navigation or reducing time-on-task stops being just a usability metric. It becomes the difference between someone successfully completing a transaction or abandoning the process entirely. Designing for those environments taught me that clarity is not cosmetic; it is access. The second is DesignFlow Kit, the open-source AI-powered workflow automation product I built from the ground up. It brought together everything I had learned from years of leading design across fintech and enterprise environments. I created it to solve a problem many existing tools were ignoring: maintaining alignment between research insights, design decisions, engineering handovers, and governance as products evolve over time. What makes it meaningful to me is its openness. Anyone can adapt it to their own workflows, operational realities, and governance needs rather than being forced into assumptions embedded in closed systems. In many ways, it is the broadest-reaching contribution I have made because it was designed to be extended beyond my own environment.

Experiences that most shaped your thinking about technology and design?
My Physics degree at Covenant University taught me to ask what is actually happening underneath the interface. That habit of refusing to take any system’s output at face value shaped the way I approach user research and product thinking. It is also why I remain cautious with AI. The polish of an output can easily hide weak assumptions, missing context, or flawed reasoning underneath. The second was my UX work at Cyberspace Limited and Interswitch Group, designing for underbanked users. Watching someone complete a transaction over USSD on a feature phone completely reset my understanding of what “good design” meant. Elegance was not the goal, access was. That single insight has shaped almost every design decision I have made since then. The third was realising that systems do not fail all at once. Whether in design, governance, or technology, breakdown usually happens gradually through small inconsistencies, weak assumptions, and disconnected decisions that compound over time. That understanding pushed me toward systems thinking — not just designing screens, but thinking deeply about how research, behaviour, infrastructure, trust, and delivery all connect together. It is ultimately what led me to build DesignFlow Kit and why I care so much about creating products that remain coherent as they scale.

One principle to guide the future of AI-driven product design?
AI should expand the agency of the user, not replace it. This is something I feel strongly as a designer. Most current AI design quietly does the opposite — autocomplete narrows what you’d have said, recommendations hide ninety-seven options to surface three, and “smart” defaults remove decisions users didn’t know they were giving up. At population scale, that compression of agency is a bigger long-term risk than hallucination or bias, and it’s a design problem more than a technical one. In practice the principle means: default to augmentation rather than automation, surface uncertainty honestly instead of performing confidence, and keep the user as the protagonist of their own decision — clinician, trader, customer, whoever they are. Products that take agency from users may win on engagement metrics in the short term, but over a generation they erode the very judgement they were supposed to support. As designers, we’re the people closest to that trade-off, which is why we have to be the ones holding the line.

One mistake some companies are making with AI that they’ll need to correct?
From a designer’s perspective, the mistake is treating AI as a feature when they should be treating it as a system. I’ve watched this play out across every sector I’ve worked in. The pattern is the same: a model gets wrapped in an API, dropped into an existing product surface, and announced as an AI capability — usually without anyone asking what it does to the user’s flow, their trust, or their sense of control. What’s missing is everything around the model: data lineage, governance, human override, drift monitoring, feedback loops, and a UX layer that helps the user understand what the AI is doing and why. The result is AI that looks impressive in a demo and underperforms in production, with no instrumentation to understand why when it fails — and a user experience that quietly erodes trust. The correction is treating AI governance and AI UX as first-class disciplines alongside engineering and security. Every previous wave — web, mobile, cloud — went through this same arc from feature-first to systems thinking. AI is mid-arc right now. The companies that move early will be the ones still trusted with AI in five years.