Hi all 
I’m building an autonomous AI maintenance assistant that works over WhatsApp (natural conversation with a field manager) and operates on structured data in monday.com (work orders, assets, vendors). The agent needs to understand intent, manage stateful workflows (new issue → scheduling → treatment → follow-up), and safely read/write to monday via tools/MCP.
I’d love advice on:
-
Whether this is best modeled as one main agent with a strong system prompt or a LangGraph-style workflow with nodes.
-
Any similar agent architectures you’ve seen (MCP-based or tool-written) that are good references.
-
Extra tips for WhatsApp-based agents (latency, message chunking, confirmations, UX pitfalls).
Thanks! 