OpenAI Codex Desktop App Streamlines R Package Modernisation

OpenAI has launched a dedicated desktop application for Codex, its AI-powered code generation tool, aiming to further streamline and automate software development tasks. The rollout of the Codex desktop app offers an interface designed to run multiple AI agents in parallel, maintain code isolation through built-in worktrees, and extend functionality with skills and scheduled automations. While Codex as a technology has been available for some time, this new application signifies OpenAI’s effort to keep pace with alternatives, such as Anthropic’s Claude Code, which have recently gained attention for similar capabilities.

A core feature of Codex’s new app is its “plan” mode, which enables users to decompose complex software tasks into manageable subtasks, incorporating user preferences and reducing the risk of errors. Users can also create reusable skills to integrate external services or schedule recurring background operations.

To showcase Codex’s utility, a demonstration was provided updating the qqman R package—a well-known package for generating Manhattan plots from Genome-Wide Association Studies (GWAS) data. The package, originally implemented in base R for performance reasons, had been the subject of numerous feature requests, particularly concerning integration with ggplot2, a widely used data visualisation library for R. Using the Codex desktop app in plan mode, the author efficiently refactored the codebase, switching visualization components to leverage ggplot2 and its extension ggrepel for clearer point labeling. The refactored package produced updated plots, more flexible column handling, improved speed, and updated documentation, completing much of this work in a matter of minutes.

Although the updated package will not be maintained further, this demonstration illustrates how AI-powered coding assistants can address long-standing software development bottlenecks, including modernising legacy packages and responding to unaddressed feature requests. Codex’s rapid execution in this context demonstrates the practical value of AI productivity tools for researchers and developers, particularly those managing complex, open-source scientific software in R or Python.

Broader commentary from AI practitioners, echoed in the article, emphasises that while AI coding tools face criticism and their limitations are real, their transformative impact on software development is undeniable.

Reference: r-bloggers.com