AI
Cursor and Google launch two-way voice for coding agents as industry backs real-time conversation
Major AI coding platforms are shifting toward interactive voice interfaces that keep developers talking to agents throughout task execution, moving beyond one-way dictation.
1/6
The New Stack
The AI coding landscape is embracing real-time conversation. On the same day in November, both Cursor and Google rolled out two-way voice capabilities that let developers communicate with their coding agents continuously—asking questions, providing feedback, and receiving spoken updates without touching the keyboard.
Voice as a collaboration mode#
Cursor's new Voice Chat with Agents runs on Grok Voice and works across web, iOS, and desktop platforms. Google's offering, Voice mode in Antigravity, integrates Gemini Live into its desktop application and command-line interface.
The motivation behind these rollouts centers on a simple observation: engineering teams naturally problem-solve out loud. Nick Van der Auwermeulen, a developer relations engineer at Google, captures the shift: "Some of the best engineering happens out loud." Traditional agent workflows force developers back to the keyboard repeatedly—whenever they need approval, spot a missed requirement, or want to change direction mid-task.
By keeping a live voice channel open, both platforms let developers stay engaged without interruption. Google's version even allows interrupting the agent mid-sentence to course-correct or ask clarifying questions.
How the platforms differ#
Antigravity's desktop app and terminal client both support voice mode, with users clicking a waveform button to initiate calls. The agent responds aloud, continues working, and delivers a written summary with code changes once complete. The CLI requires a simple /voice live command.
Cursor emphasizes breadth: voice chat reaches users on web browsers and mobile devices alongside desktop. The platform builds on earlier dictation work added in October 2025, where agents already offered brief spoken progress updates.
The broader industry shift#
Cursor and Google aren't pioneers here. OpenAI introduced two-way voice in its Codex desktop app and ChatGPT desktop client earlier this year, and the Codex CLI has supported live voice conversations since spring. Anthropic's Claude offers bidirectional voice in its main app but limits Claude Code's voice feature to dictation only—the /voice command still converts speech to text, with replies returning in written form.

