A developer shares quick, personal impressions after using Codex more than Claude for a week. While both assistants were kept on par with plugins and skills, Claude had more skills initially, requiring manual porting to Codex.

When debugging urgently, the author still opened Claude due to familiarity, despite Codex not necessarily being inferior. Codex produced Ruby on Rails changes with fewer comments, which was appreciated.

Codex's agent harness output felt more technical, like "Data from Star Trek," whereas Claude felt like a collaborative colleague. Codex completed changes faster but spent more time finishing pull requests with tests and reviews, resulting in no overall time savings.

Codex created simpler code architectures compared to Claude, which tended to add abstractions and Sorbet signatures. However, Claude's more complex solutions handled edge cases better. Codex made branching mistakes, creating tangled PRs, while Claude better understood intentions to sync work branches.

For Jira and Atlassian integration, Claude was more adaptive, while Codex struggled with authentication flows. With MCPs, the Codex CLI approach was preferred for its explicit login prompts. The key takeaway: Claude tries to anticipate and exceed expectations, while Codex is a literal companion that does exactly what's asked without overreach.