Willison used Codex voice mode for most of a Django newsletter archive, then reviewed the pull request and finished the project with typed prompts.
Developer Simon Willison reported on October 9 that he used OpenAI Codex voice mode with GPT-6 Astra High to create a Newsletters section for his Django blog. The resulting index brings together his free weekly Substack posts and monthly sponsor updates, with links to archives organized by year.
Willison started in the Codex tab of the ChatGPT desktop app, connected it to his local development environment and launched a server that provided a browser preview. He then opened a separate voice chat and described the feature while cooking dinner. He estimated that this voice-driven phase lasted about half an hour.
During the session, the agent produced a Django data model and migration, administrative configuration, views and templates. It also implemented four import mechanisms: recent Substack entries through RSS, the remaining Substack material through an undocumented API, published monthly newsletters from a public GitHub repository, and the latest private sponsor newsletter from a private repository.
The new content appeared in daily and monthly archives and in site search, but not on tag pages or the homepage. Weekly editions linked to Substack, while published monthly newsletters received their own pages. According to Willison, GPT-6 Astra also designed the section and adjusted it in response to his spoken feedback as he checked the local preview.
The project was not completed entirely through voice commands. After Codex created a branch and opened a pull request, Willison reviewed the code in GitHub and found that one importer used Git through a subprocess. He considered an API more suitable for retrieving data from the private repository, so he switched to typed prompts, replaced the import mechanism and refined the public pages. That final stage took roughly another half hour before deployment.
This account suggests a practical boundary for voice-based coding: it helped Willison define the feature, clarify its behavior and follow progress while doing another task, aided by the visual preview. He still found typing more efficient for precise changes, error messages, examples and code references. This was one developer’s experience, not a comparative Codex test or evidence that other projects would produce similar results.
Sources
Originally published by AI Feed: Simon Willison Builds a Django Newsletter Archive Using Voice Commands.
Top comments (0)