Keep pulling the thread on Shiv and Mikal.
Shiv believes that innovating on business models is the most critical task for AI in healthcare at the present moment.
Shiv asserts that providing sufficient context to clinical AI models is critical, as sparse context can result in answers that are "deadly wrong."
Mikal, a software engineer at Abridge, developed an agentic patient intake system called Previsit AI while he was still a practicing clinician.
Mikal claims that an LLM coded the MVP for his hackathon project in approximately 40 minutes, complete with test coverage and documentation.
Mikal believes that Anthropic's Opus 4.5 and 4.6 models, released around December 2023, represent a significant acceleration in the coding and understanding capabilities of large language models.
Mikal quit his job as a full-time interventional cardiologist in Belgium to join Abridge.
Mikal asserts that while AI tools enable rapid prototyping by clinicians, shipping production-grade solutions in healthcare requires professional engineers to ensure safety and prevent errors.
Shiv observes that health systems are becoming more deliberate in deciding when to partner with vendors versus investing in their own internal innovation.
Shiv believes that for high-stakes problems in healthcare where 80% accuracy is insufficient, there is a significant market opportunity for dedicated companies to invest heavily in achieving near-perfect solutions.
Shiv categorizes the evolution of AI companies into three distinct eras: post-transformer/pre-LLM, post-LLM/pre-Agent, and the current post-Agent stage.
Abridge's VP of Product, Dany, has implemented a new internal process where building prototypes replaces the traditional writing of Product Requirement Documents (PRDs).
Mikal cites the startup PMcardio, which uses machine learning models to detect closed coronary arteries from ECGs that are not detectable by standard criteria.