Cursor and Google have joined OpenAI in giving AI coding agents a voice. The two companies have introduced two-way voice features that let developers speak to their agents and receive spoken responses in return.
Cursor launched ‘Voice Chat with Agents,’ a feature powered by xAI’s Grok Voice. Developers can speak instructions to the agent and hear its replies. The company positions the feature as a way to keep working without breaking focus or switching between screens.
OpenAI has also added voice support to Codex, its AI coding agent. Developers can start a voice conversation with a Codex task in the ChatGPT desktop app, speak instructions, interrupt responses and ask follow-up questions while the agent works. Voice is also supported in Codex CLI, allowing developers to speak to Codex and hear its responses directly in the terminal.
Anthropic’s Claude Code has not joined this shift. The tool supports a /voice command, but it works only as speech-to-text dictation. Developers can speak a prompt, and Claude Code converts it into text.
The agent then responds in writing. It does not speak back, join a live call or provide audible updates while working. This keeps Claude Code focused on a text-first workflow, where developers review prompts, code changes and explanations on screen.
Anthropic does offer two-way voice in the broader Claude app. However, it has not extended that conversational experience to Claude Code. The company has not announced a timeline for adding spoken responses to its coding tool.
Google added a similar capability to Antigravity, its AI coding platform. The feature uses Gemini Live and is available through the desktop app and command-line interface. Developers can start a voice session with the ‘voice live command,’ issue instructions and receive spoken updates from the agent.
The move marks a shift in how developers may interact with AI coding tools. Instead of typing a prompt, waiting for output and reading a response, developers can now hold a conversation with an agent while it works on code.
The feature may be most useful during iterative work. Developers can describe a change, listen to the agent’s plan and follow up while the code is being generated. This could make AI agents feel less like command-line tools and more like collaborators.
But voice is unlikely to replace written output. Code reviews, diffs, logs and error messages still need visual inspection. Developers may prefer to speak during planning and execution, then read the final result carefully.
Also Read: Anthropic Unveils New Claude Models as AI Self-Improvement Debate Grows