The conversation highlights a significant trend where advanced AI coding assistants are empowering clinicians with deep domain expertise to become proficient software builders. Mikal's experience of building an agentic patient intake system and winning a major hackathon as a practicing cardiologist is a prime example of this new archetype.
Shiv argues that the most important task for AI in healthcare today is to innovate on business models. The technology's potential cannot be fully realized within the existing adversarial framework of payers and providers; instead, it must be used to create new, collaborative models centered on patient value, such as streamlining prior authorization.
The speakers distinguish between two primary fronts for AI adoption. Administrative AI, which 'liberates physicians from the keyboard,' is poised for rapid adoption due to its clear ROI and lower regulatory hurdles. In contrast, core clinical AI for diagnostics and treatment will be adopted more slowly, requiring rigorous validation and adherence to evidence-based medicine.
The discussion points to a shift from AI as a passive tool to AI as an active, agentic partner. This is demonstrated by Mikal using an LLM to not only code but also manage server infrastructure, and by Abridge's internal adoption of 'prototypes as the new PRDs' and personal AI assistants for tasks like research and email prioritization.
Keep pulling the thread on Shiv and Mikal.