Reading room
Voice-driven coding agents can build real features, but typing still wins for precision
Simon Willison describes building and deploying a newsletter feature for his Django blog largely through a voice conversation with a coding agent while cooking. The agent created a model, migration, imports, archive pages and search integration, then Willison reviewed the resulting pull request and replaced a Git-subprocess import with an API-based implementation. He found voice useful for multitasking and high-level iteration, but typing remained more efficient for code details, examples, errors and final corrections.
Editorial summary · Publisher article
What matters
- The workflow combined voice prompts, a visual local preview and a later typed code review.
- The agent implemented database, import, archive and search changes across the site.
- A production-relevant import design was changed during human review.
- The author sees voice coding as valuable for multitasking rather than a daily replacement for typing.
Context & caveats
- This is one developer’s experience with a specific model, application and project; it is not a controlled evaluation.
- The account includes the author’s own judgments about the model’s performance and does not independently verify every implementation detail.
Why this made the edition
This is a detailed first-person account of building and reviewing a real Django feature with voice-driven coding assistance, including workflow benefits, limitations and a concrete correction before deployment.
Source material