Simon Willison’s discussion of Sean Lynch’s comment on MCP underscores a potential shift in how authentication is handled in AI systems. While isolating auth flows from LLM context windows could streamline API interactions, it’s unclear whether this approach addresses a widespread pain point or merely solves a narrow technical issue. Critics might argue that this innovation, while clever, could be overshadowed by more pressing challenges in AI infrastructure. The broader question is whether MCP’s focus on authentication represents a meaningful step forward or simply a minor tweak in a rapidly evolving field.