OpenAI realized that developers often need to run multiple variants of the same task on AI and programming simultaneously.
I’ve already mentioned this in an article about managing large projects with many dependencies.
Therefore, now in Codex, you can run up to four variations of the same task at the same time and then select the most suitable answer.
Of course, this wouldn’t work with people—they lack the ability to test several solutions in parallel.