Quick impressions: A week of using Codex more than Claude

On macOS, Codex TUI (gpt-5.6-sol xhigh) produced fewer Ruby comments, more terse output, and simpler architectures than Claude Code TUI (opus-5 xhigh), but it mishandled branch rebases and required manual Jira login. Claude supplied more built-in skills, Sorbet signatures, and proactive interaction, though its broader session handling added overhead. Both met the requirement; Claude’s code was more complex and case-rich, while Codex’s speed advantage disappeared after extensive testing.