Dylan Fox attributes AssemblyAI's growth to three macro trends: improved context-aware voice models, a rapidly evolving AI infrastructure ecosystem, and the rise of coding agents, which have expanded their total addressable market by 100x. He emphasizes that data quality and mixture drive 75% of model performance, making real-world evals more important than open-source benchmarks. Fox discusses the technical hurdles in humanoid robotics, such as speaker disambiguation, and predicts voice will become a standard interface within 18 months. He also addresses the sovereign AI problem, noting that customers are returning from self-hosted models due to maintenance difficulties. On the UX front, he debates whether voice agents should disclose they are AI, arguing against deception. Finally, he highlights the upcoming surge in consumer applications powered by on-device models, which will make voice a core feature in everyday devices.