How to build an AI-Native Health Company — Dan Feng, Maven Clinic
Dan Feng outlines Maven Clinic's transformation into an AI-native company, driven by the realization that "building is cheap and arguing is expensive" in the AI era. This shift has reshaped planning to short, iterative sprints, revolutionized software development with AI coding tools, forced adaptive changes in code review processes, and necessitated a nuanced, multi-layered approach to ensuring reliability in generative AI systems.
Healthcare’s Agent Bytecode: X12 as the Harness for AI Agents — Vasant Kearney, Onlay
Vasant Kearney presents a nuanced perspective on implementing AI in healthcare, particularly in insurance. He highlights that while AI is evolving rapidly, real-world application requires understanding that solving isolated problems doesn't equate to complex solutions. His core argument is the utility of X12 as a 'harness' for agentic AI, providing a structured, confining framework for LLMs in healthcare. This approach acknowledges that X12 defines the contract between providers and payers, and even phone calls can be seen as X12 transactions. However, he cautions that X12, or any payer data, isn't ground truth due to disparate system origins, necessitating an internal 'semi-correct' representation. Kearney advocates for a balanced "AI pilled and AI skeptical" stance, embracing AI's potential while remaining cautious about its inherent limitations, cost, and the necessity for robust system design and evaluation.
MCPs for Observability Stacks
Diana Todea, Head of Developer Relations at VictoriaMetrics, demonstrates how Model-Controller-Pair (MCP) servers enhance observability stacks for faster, smarter troubleshooting. She highlights the use of AI-driven anomaly detection, natural language querying, and customizable "skills" to investigate production issues, manage metrics, and generate alerts, showcasing practical applications with two interconnected MCP servers.
The Next Medium: Why Real-Time Interactive Video Changes Everything — Ahmed Ahres, Reactor
Ahmed Ahres argues that real-time interaction fundamentally changes the medium, not just its speed, especially in generative AI. He defines "world models" as interactive, effectively infinite, and steerable video, drawing parallels with GPS and camera viewfinders. This paradigm shift unlocks new forms of control, intelligent advertising, programmable worlds for robotics and education, and advanced live avatars. He highlights the critical infrastructure challenges of streaming pixels, managing stateful sessions, and achieving global sub-100ms latency, emphasizing that batch infrastructure is unsuitable for real-time applications. Evaluation of consistency for these models remains an unsolved problem.
Continual Learning: How AI Agents Get Better With Every Use | Arjun Karanam, Trajectory
Arjun Karanam from Trajectory discusses the "experience gap" in AI, where models excel in intelligence but lack real-world experience, advocating for continual learning. He outlines four key areas for the agent ecosystem: robust traceability including corrective actions, evaluations drawn from production traffic, harnesses that orchestrate rather than constrain, and comfort with open-weight models. Trajectory aims to provide a platform for companies to own and continuously improve their AI intelligence.
Why Deep Networks Don’t Need to Memorize Everything — Matthieu Wyart
Matthieu Wyart, a statistical physicist, argues that deep networks discover abstractions by recovering hidden data hierarchies, allowing them to escape the curse of dimensionality. He explains how this mechanism, combined with predicting latent representations instead of raw tokens, can significantly improve sample efficiency. The discussion also covers the physics of rough loss landscapes, machine creativity, diffusion models, and a theoretical framework for neural scaling laws.
