Conductor vs. native Claude Code. Same single-agent performance?

Now that people have been running it for a few months, what's the verdict on single-agent performance vs native terminal Claude Code?

One thing I noticed is that Conductor bundles its own (pinned) version of Claude Code and Codex instead of using the latest one from your system. Does that bother anyone, or is it a non-issue?