Developer shares personal impressions after a week of using Codex more than Claude for coding tasks

Comparative Usage of AI Coding Assistants

A developer conducted a personal comparison of Codex and Claude, two AI coding assistants, over a week. The goal was to evaluate their performance and usability in daily development tasks, particularly in Ruby/Ruby on Rails environments. This followed a period where Claude had been the primary tool, accumulating more custom skills.

Code Output and Architectural Differences

Codex generated Ruby/Ruby on Rails code with fewer comments, which the developer found preferable. The output from Codex was described as more "technical," contrasting with Claude's more conversational style. In terms of code architecture, Codex tended to produce simpler solutions, while Claude often introduced more abstractions, concepts, Sorbet signatures, and type aliases. Although Claude's code was more complex, it handled more edge cases.

Workflow Integration and Speed

Despite initial impressions of Codex being faster for making main changes, the overall time to complete a pull request was similar to Claude due to extensive testing and review processes. The developer also noted a preference for opening multiple focused sessions with Codex, differing from previous habits of using larger, single sessions with Claude. When debugging urgent issues, the developer instinctively reverted to Claude due to familiarity.

Challenges and Specific Behaviors

Codex exhibited some issues with Git branching, such as creating incorrect branch dependencies (e.g., branch A targeting branch B which targets main). This led to large, unmanageable pull requests when attempting to rebase, requiring explicit instructions to rebase only with the intended target. Claude, in contrast, demonstrated a better understanding of the developer's intention to keep branches synchronized.