Right now I have three agents writing code for other agents.
1. An agent session where an agent is writing docs for other future internal operations agents to follow
2. An agent session where an agent is writing an LLMS.txt which will help user's own agents drive our public endpoints so the users don't have to do it themselves
3. An agent session that is writing a testing harness so that other coding agents can end-to-end test their coding
Even weirder, with #2, the agent is iterating by spinning up its own subagents in clean room environments fed only the LLMS it wrote, so that it can see where those agents fail and update the LLMS. I'm not even consulted on this.
This is some crazy recursive/exponent looking timeline we're in.
