Asynchronous Delegation

Asynchronous delegation splits a conversation into two jobs: one that keeps the dialogue moving in real time, and one that thinks deeply to produce an answer. By assigning these to different places, the thinking time never blocks the flow of conversation.

Think of a call center. The agent who answers the phone keeps talking with the customer, and when a tough question comes up, passes a note to a specialist in the next room. While the specialist digs up the information, the agent keeps the call alive by saying things like "Sure, let me check that for you." Once the specialist's answer arrives, the agent picks it up smoothly and relays it back.

In voice AI, this same split happens between a model that exchanges speech quickly and a heavier model that handles search or complex reasoning, with each running on its own separate path. Even if the search takes longer, the voice conversation itself never stalls, so the user can keep talking without any sense of waiting.

As it appears in the article: "Technically, this is called asynchronous delegation. Audio moves only on a dedicated high-speed path between the client and the voice model, while search, tool calls, and deep reasoning are handled on a separate path."

Correcting a misconception: this doesn't mean the voice model itself carelessly brushes off hard problems. Rather, the model that handles the conversation and the model that reasons deeply run on physically separate paths, by design, so that delays on the slower side never affect the response speed of the faster side.