Claude can directly drive Codex or Codex can drive Claude. Both already produce logs. It's unclear what value this intermediary brings.
It's also unclear what conversing agents are useful for other than wasting money, energy and water.
It's hard enough to get the same model to be consistent around it's vision let alone multiple of them.
I'm building an EMR and the other day asked Claude what a decent model would look like for capturing wound orders. Then, I took the output, started a new session and asked the new session to critique that model and the response made me want to pull my hair out. It blasted the model from it's former self and suggested making a ton of updates.
I'm sure more scoped tasks would fair better, but it was pretty frustrating.