The topic of discussion is the use of artificial intelligence in high-stakes fields such as medicine, where a confidently wrong answer can have serious consequences. Sumit Gundawar, a London-based software engineer, argues that in such fields, earning trust is the real engineering challenge, rather than just building the model. Gundawar works on the clinical platform for a UK longevity and aesthetic-medicine clinic, where patient safety is a top priority. He emphasizes the importance of grounding, refusal logic, and human-in-the-loop design in AI systems to prevent patient-safety events. The conversation also touches on the concept of hallucinations in AI, which can be particularly problematic in medical contexts. Gundawar demonstrates an assistant that refuses to answer when it cannot back up its claim, highlighting the need for transparency and accountability in AI decision-making. The discussion also covers various tools and platforms, including Anthropic, OpenAI, and LangChain, which are being used to develop more reliable and trustworthy AI systems. Additionally, the conversation mentions regulatory frameworks such as the EU AI Act and HIPAA, which aim to ensure the safe and responsible development of AI in high-stakes fields. The importance of human oversight and review in AI decision-making is also emphasized, particularly in fields where the consequences of error can be severe. Overall, the discussion highlights the need for a more nuanced and multidisciplinary approach to developing AI systems that can be trusted to make decisions in high-stakes contexts.