How long until LLMs are prompting LLMs to write a response to their user query?

Actually, this happens already in a modular way AFAIK…

This already happens with -- e.g. -- Claude Code spawnining parallel agents and then collating their results.