Skip to content
Stack Courier SC markSCSTACKCOURIER
Google ADK empowers devs to build responsive voice agents using non-blocking Python tools and gemini-3.8-live
Depth
14.4 secondsGoogle claims switching to streaming tools cuts time to first words from 14.4 seconds to 1.2 seconds in benchmark tests · reported
1.2 secondsGoogle claims switching to streaming tools cuts time to first words from 14.4 seconds to 1.2 seconds in benchmark tests · reported
1 outlet · newest source 366h ago
Agent builders

Google ADK empowers devs to build responsive voice agents using non-blocking Python tools and gemini-3.8-live

The open-source Agent Development Kit lets developers build responsive voice agents using non-blocking Python tools and gemini-3.8-live streaming capabilities.

Key points

  • Google claims switching to streaming tools cuts time to first words from 14.4 seconds to 1.2 seconds in benchmark tests.
  • Developers can use yield statements in AsyncGenerator functions to create non-blocking tool calls that provide immediate status acknowledgments.
  • Response scheduling options like WHEN_IDLE and INTERRUPT give precise control over when tool results enter the conversation flow.

Google reported that its open-source Agent Development Kit, combined with gemini-3.8-live, allows developers to eliminate dead air in voice agents by implementing non-blocking tools in Python.

Evidence: reportedStep by Step guide: Build a Voice Agent That Never Goes Silent using ADK

aifleet-google-dev

What was released or claimed

In a blog post published on the Google Developers blog, Google detailed how the ADK and gemini-3.8-live enable responsive voice agents through streaming tools and non-blocking function calls. The company stated that in a benchmark sales-report workflow, switching from a blocking tool to a streaming tool reduced the time to the agent's first words from 14.4 seconds to 1.2 seconds.

Evidence: confirmedStep by Step guide: Build a Voice Agent That Never Goes Silent using ADK

aifleet-google-dev

Evidence: reportedStep by Step guide: Build a Voice Agent That Never Goes Silent using ADK

aifleet-google-dev

Who it affects

The update targets developers building live voice conversations using the Gemini Live API. It specifically assists those who need to keep agents conversational while slow backend queries, such as database lookups or RAG pipelines, execute in the background.

What is still unknown

The source material does not specify the hardware requirements for running these tools or the exact latency thresholds for other model versions. It also does not detail the cost implications of using streaming tools versus standard blocking calls.

The takeaway

Standard async def functions create blocking tool calls that halt conversation until the function returns a value. By using streaming tools with yield statements, developers can send partial updates upstream, allowing the model to paraphrase progress while the tool executes.

What changes for users

Data science teams can now implement non-blocking voice tools natively in Python without wiring up custom client-side WebSocket event loops. Developers can choose between streaming tools for multi-stage workflows or background execution patterns. They can also configure response scheduling behaviors such as WHEN_IDLE, INTERRUPT, or SILENT to control how tool results are delivered into the conversation.

Why it matters

  • Teams can reduce perceived latency in voice interfaces by providing immediate audio acknowledgments while backend processes run.
  • The ADK simplifies asynchronous tool state management, removing the need for custom WebSocket client-side logic.
  • Developers gain granular control over conversation flow by scheduling when results interrupt or update the dialogue context.
  • The approach addresses the specific UX issue of dead air in live voice sessions, which previously caused users to hang up.

Topics: Google, ADK, voice agents, gemini-3.8-live, Python

Updated

Numbers in context

14.4 secondsreported1.2 secondsreportedCircle area is proportional to the magnitude; labels retain signs. Figures may describe different measures. Read their contexts above.

Ask the article

Local keyword search of this article; no generated answers.

Which of these is confirmed?

Check the answer

In a blog post published on the Google Developers blog, Google detailed how the ADK and gemini-3.8-live enable responsive voice agents through streaming tools and non-blocking function calls.

The matched primary source and announcement wording support the confirmed label.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Questions readers ask

How do streaming tools differ from standard tools in ADK?

Streaming tools use yield statements to send partial updates, allowing the agent to speak immediately, while standard tools block the conversation until the function returns a single value.

What response scheduling options are available?

Developers can configure WHEN_IDLE to wait for a speaking pause, INTERRUPT to speak immediately, or SILENT to update context without speaking.

Does this require custom WebSocket code?

No, the ADK and gemini-3.8-live allow developers to implement non-blocking voice tools natively in Python without custom client-side WebSocket event loops.

Connected stories

This storyGoogle ships A2UI 1.0 protocol for native generative UI agentsGoogle ships A2UI 1.0 protocol fGoogle A2UI Launches Declarative UI Standard for CopilotKit and Agent FrameworksGoogle A2UI Launches Declarative

Reading receipt

  • Google claims switching to streaming tools cuts time to first words from 14.4 seconds to 1.2 seconds in benchmark tests.
  • Developers can use yield statements in AsyncGenerator functions to create non-blocking tool calls that provide immediate status acknowledgments.
  • Response scheduling options like WHEN_IDLE and INTERRUPT give precise control over when tool results enter the conversation flow.
Time spent: 0:00 · Depth: Full

Sources

  1. Google Developers blog

Related stories