Developer builds blog feature entirely via voice using Codex voice mode

Simon Willison reports shipping a new Newsletters index page for his blog using OpenAI’s Codex voice mode in the ChatGPT desktop app. He conducted the session against a local Django development environment while cooking dinner.

Willison initiated the session by typing a command to start the dev server and open the site in a browser, then switched to voice mode to describe the desired feature. He used GPT-6 Astra High to design a new Django model and migration, write view code and templates, and implement import scripts to populate the database from multiple sources.

Over roughly 30 minutes, the model produced a working feature: a new model and migration for newsletters, Django Admin configuration, four import scripts (recent Substack items via RSS, older Substack items via an undocumented API, monthly newsletters from a public GitHub repository, and a private sponsors-only newsletter), and the /newsletters/ and /newsletters/2026/ archive pages. The model also integrated the new content into day and month archive pages and the site search engine, while excluding it from tag pages and the homepage.

Willison reviewed the generated code in a GitHub pull request and switched to typing to adjust one import script from Git-subprocess to an API-based approach, citing the need for a private repository access token. He merged the changes after an additional 30 minutes of typed prompting and deployed the feature to production.

Willison characterizes voice-driven coding as effective for multi-tasking and early-stage design, but less efficient than typing for detailed code review and error handling. He notes the combination of voice input with a live visual preview and the ability to paste in code snippets or error messages as a powerful interaction model for coding agents.