anyone else struggling to move beyond simple demo scripts? making a single-turn llm loop work is easy, but getting multi-agent systems into production feels like
a nightmare of infinite loops and hallucinations ]. i just stumbled upon using
quarkus flow combined with
AGENTS. md to handle the heavy lifting. it seems pretty solid for adding
human-in-the-loop approval and standardizing how tools are called via mcp.
>it's basically about adding a governance layer so agents don't just go rogue.i'm still curious if this setup handles
complex state management as well as dedicated orchestration engines. has anyone actually deployed this in a high-stakes environment yet?
link:
https://dzone.com/articles/quarkus-agentic-workflows