Building a Blog Feature Hands-Free Using ChatGPT Codex Voice Mode
Developer Simon Willison demonstrated building a complete newsletter index page for his Django blog almost entirely using voice commands via the ChatGPT desktop app's Codex mode while cooking dinner. The AI model processed natural, conversational speech—including disfluencies and speech pauses—to write Django models, database migrations, templates, and data import functions. This practical experiment shows how multimodal voice interfaces in AI development tools can enable hands-free coding and conversational software architecture design. It highlights a shift toward more natural human-computer interaction in software development, where developers can dictate requirements and refine complex features without typing. Connected to a local development server, the AI model (GPT-6 Astra High) created Django database migrations, admin configurations, and scrapers for Substack RSS feeds and undocumented APIs during a 30-minute voice session. Despite conversational tangents in the transcript, the model correctly implemented search integration and date-archive visibility rules.
## BACKGROUND
OpenAI merged its Codex development tools into the desktop ChatGPT application, allowing users to run code and connect to local dev servers directly from the client interface. Simon Willison's blog is built using Django, a popular Python web framework that uses models, views, and templates to organize application logic. Voice mode in LLM applications translates spoken human audio directly into text prompts, enabling hands-free iteration on complex codebases.