Model Labs Won't Kill App Startups | Herbie Bradley

AI agent swarms are shifting from static, database-reliant models toward dynamic, in-weight learning architectures that promise greater coordination and autonomy. While current RAG-based systems effectively support individual workers, automating entire firms remains challenging due to the complexity of relational tasks like negotiation and human-centric management. Robotics progress, coupled with foundation models, suggests a path toward self-replicating lunar factories and specialized manufacturing, potentially enabling von Neumann probes within the next 15 years. Despite the dominance of large labs, application-layer startups maintain a competitive edge through vertical integration and the ability to synthesize multiple model providers. These advancements reflect a broader transition where AI increasingly delivers measurable economic outcomes, though real-world inertia and social bottlenecks continue to shape the pace of technological deployment and infrastructure development.