Over the past two years, I have spent a significant amount of time thinking about a question that continues to shape almost every discussion around artificial intelligence in healthcare: why does so much of the technology appear so promising, yet struggle to translate meaningfully into everyday clinical practice?
This question did not emerge from academic theory or observation from a distance. It came from practical experience. As a pharmacist and founder building within healthcare, I have had the opportunity to observe firsthand the growing excitement surrounding artificial intelligence, while also witnessing the realities of how healthcare actually functions in practice. What became increasingly clear to me was that there appeared to be a widening gap between how AI was being discussed and how healthcare professionals actually work.
Healthcare is often spoken about as though it were a straightforward environment for technological implementation. In reality, it is anything but. Clinical practice exists within a world of interruptions, uncertainty, competing priorities, regulation, governance, and professional accountability. Decisions are rarely linear, consultations rarely unfold exactly as planned, and no patient presents as a textbook example. A healthcare professional must constantly balance evidence, judgement, risk, patient preference, and practical constraints, often under considerable pressure.
Yet many of the conversations surrounding healthcare AI seemed to focus almost exclusively on capability. How advanced is the model? How quickly can it process information? Can it outperform clinicians in specific tasks? While these questions are important, I increasingly felt they were not addressing the issue that mattered most. In healthcare, the success of artificial intelligence will not be determined solely by how intelligent a system appears in isolation, but by whether it can function safely, responsibly, and meaningfully within the realities of clinical workflows.
This distinction became increasingly important to me as I continued developing solutions through PharmBot AI and AIVAe. The more time I spent working within pharmacy workflows and structured consultations, the more convinced I became that many failures in healthcare AI are not necessarily failures of technology. Rather, they are failures of implementation and context. Too often, solutions are designed around idealised environments rather than the realities of practice, overlooking the complexity of clinical responsibility and underestimating the importance of governance, accountability, and workflow integration.
This growing conviction ultimately became the reason for writing AI Isn’t Failing in Healthcare — It’s Built in the Wrong Place.
The purpose of this book was not to position myself as having definitive answers, nor to dismiss the extraordinary progress that has been made within artificial intelligence. On the contrary, I remain deeply optimistic about the role AI can play in transforming healthcare over the coming decade. However, I have increasingly come to believe that meaningful transformation will not occur simply because models become more sophisticated. Instead, it will depend on whether we are able to design systems that complement the realities of clinical practice, support rather than disrupt professional workflows, and operate within frameworks that healthcare professionals can trust.
At its core, the book argues for a relatively simple but, I believe, increasingly important idea: healthcare AI succeeds not when it attempts to replace professional judgement, but when it is designed around the environments in which healthcare professionals already operate. Technology in healthcare cannot exist independently of workflow, governance, safety, or accountability. If these factors are treated as secondary considerations, even the most technically advanced systems are unlikely to achieve meaningful adoption or sustained impact.
Writing this book was also, in many ways, a reflection of the personal journey of building within healthcare innovation. Like many founders working in difficult sectors, the process has included setbacks, uncertainty, rejection, and periods where progress has felt slower than expected. Building something meaningful in healthcare rarely follows a straight line, particularly when attempting to navigate clinical complexity, regulatory considerations, and limited resources simultaneously. Nevertheless, throughout this process, my belief in the importance of the problem itself has remained unchanged.
Healthcare deserves technologies that genuinely understand the environments in which they are expected to operate. If artificial intelligence is to fulfil its potential, it must move beyond impressive demonstrations and begin solving problems in ways that align with the realities of frontline care.
This book is my attempt to contribute to that conversation. Not as a conclusion, but as a starting point for asking better questions about how we build, implement, and ultimately trust artificial intelligence within healthcare.
If it encourages reflection, constructive debate, or even a different way of thinking about healthcare AI, then I believe it will have served its purpose.