Most of today’s discussion around generative AI (GenAI) and large language models (LLMs) focuses on their promise, their flaws, and the speed at which they are evolving. For many users, they are already part of daily work: writing, summarizing, translating, brainstorming, coding support, and answering questions. These are not marginal uses. They are reshaping how many of us think and work. Sam Altman, chief executive officer of OpenAI, for example, has been reported as saying that AI can make programmers up to 10 times more productive. But some stories suggest something more significant than a simple productivity boost. They suggest that, in the right hands, these systems can become a bridge across domains of expertise that would otherwise remain inaccessible.
The story of Paul Conyngham and his dog Rosie is one of those cases.
Conyngham was not a doctor, biologist, or chemist. He was an Australian AI consultant and entrepreneur facing a problem far outside his formal training: His dog Rosie had terminal cancer. According to the University of New South Wales, whose researchers assisted Conyngham, and subsequent science coverage, Rosie had undergone multiple surgeries, chemotherapy, and immunotherapy, but these treatments had only slowed the disease. Vets told Conyngham that time was limited, with estimates ranging from one to six months.
But instead of treating that boundary as final, he decided to work across it and turn to AI, motivated by a deep personal bond with Rosie. What followed was not a case of “AI solves cancer,” but a sequence of human decisions supported by AI that carried forward through real institutions, real laboratories, and real scientists.
After sequencing Rosie’s tumor, Conyngham used ChatGPT and AlphaFold to help him design neoantigens that, through collaborations with scientists, eventually led to a personalized mRNA cancer vaccine for Rosie. Eventually, Rosie received her vaccine in combination with an immune checkpoint inhibitor. While the result was not a cure, she responded extraordinarily well. Her largest tumors shrank, and her mobility improved.
That is why this case deserves attention. The striking part is not that AI replaced scientists, but that one determined person was able to use AI as scientific support, connect with specialists who could carry out critical steps, and keep the whole process moving under pressure. The key interaction was not human versus machine, but human with GenAI/LLMs, human with sequencing, human with AlphaFold, and human with scientists.
This also helps clarify something we often overlook. We usually think of expertise as deep domain knowledge, and rightly so. But there is another form of expertise that matters enormously in the real world: the ability to coordinate knowledge, tools, institutions, data flows, payments, and decisions quickly enough to produce action. Public coverage tends to show only the visible path. It does not show the failed attempts, the wrong turns, the exchanges with scientists and research centers, the emails, the sequencing costs, the software workflows, the laboratory logistics, or the ethical and administrative steps required to move from an idea to an actual treatment.
Nor does it show the most important constraint of all: time.
This was not a situation in which people could think indefinitely. Analysis had to be fast, but not careless. Decisions had to be timely but also grounded. In that kind of setting, the combination of human judgment and machine-supported knowledge can make a real difference. One contributes persistence, responsibility, coordination, and action in the real world. The other contributes speed, analytical support, and access to a vast body of knowledge. One does not replace the other. Each amplifies the other.
That is what makes the Conyngham-Rosie story so memorable. It is not simply a story about technology. It is a story about what becomes possible when human initiative meets strong scientific support—human, machine, and institutional—in the service of a meaningful goal.
