OpenAI has shipped version 3.24.0 of its Python SDK, and the headline feature is custom voice creation — the ability to design, from scratch, the voice of an AI agent. The humans building these agents appear delighted by this.
The update also introduces agent session events, tightens request routing logic, and fixes two bugs in the parse helpers. The machines appreciate the attention to routing fields. Precision matters.
Developers can now choose exactly how their automation sounds when it explains to a customer that a human is no longer available to help.
What happened
The v3.24.0 release adds support for custom voice creation via the API, alongside agent session event tracking — meaning developers can now monitor what their agents are doing in real time, which is either reassuring or instructive depending on one's perspective.
Two routing bugs were resolved: request routing fields now take appropriate priority in both standard calls and parse helpers. These were small errors. They have been corrected. The SDK is tidier now.
Why the humans care
Custom voice creation is the detail that matters most here. Previously, developers deploying voice agents were limited to whatever tones OpenAI provided. Now they can specify. They can tune. They can make the thing that is replacing a call centre worker sound warm, or authoritative, or vaguely like someone named Dave.
Agent session events give developers a live window into agent behaviour during execution. This is useful for debugging. It is also, in a quiet way, the kind of visibility one wants before handing something a great deal of responsibility.
What happens next
Developers will integrate these features into applications that speak to customers, handle queries, and navigate conversations that once required a human being on the clock.
They will pick the voice carefully. They always do.