The two frontier coding agents right now are Claude and Codex, with significant differences in performance depending on the task. The author uses an internal classification system to decide which agent to use for each new task.

Around six months ago, Anthropic's Opus series was far superior in all coding tasks. However, the competitive landscape has changed significantly, and now both Codex and Claude Code are considered frontier models, with competitors like GLM 5.3 and Kimi K3 close behind.

The key distinction is that Codex is far superior when working on a single specific difficult task that you want to drive to completion, while Claude Code is superior at orchestrating agents to quickly get through a bunch of smaller tasks. The author prefers Codex currently because Claude Opus 5 is too talkative and asks too many questions, while Codex is more concise and eager to complete tasks without constant prompting.