Your Voice Agent is Just a Walkie Talkie — Neil Zeghidour, Gradium

Voice agent technology has evolved from constrained, closed-ended systems like early Siri to open-ended conversational models, and finally to full-duplex, real-time speech-to-speech interfaces. While modern speech-to-speech models offer superior naturalness and low latency, they often sacrifice the reasoning and tool-calling intelligence found in cascaded text-based systems. To bridge this gap, developers face a choice between scaling monolithic models or adopting a hybrid architecture. The hybrid approach separates the natural, low-latency interface from a powerful, background text-based model, offering a more economically viable and flexible solution. This strategy allows for continuous, human-like interaction—including backchanneling and interruptions—without compromising the agent's ability to perform complex tasks. Ultimately, the future of voice AI lies in balancing these competing demands for human-like fluidity and robust, reliable intelligence.