Ask HN: Anyone using "Deep Agents" for production or operational tasks?

Is anyone using deep agents (LangChain or others like Claude Code, Codex) internally in their company or for real work that is non-coding?

I'm building one for a specific task, mainly to run through cron (but users can also chat/ask/provide feedback), and I'm trying to understand best practices.

Can anyone share any examples? I'd like to see something that is not a research agent, a coding agent, or a data/financial analyst agent, but something that does real work.