I’ve been experimenting with multi-agent setups lately, and the biggest headache has definitely been dealing with unexpected tool-call storms. When agents start triggering loops or redundant calls, the cost and performance impact hits fast daily agentic ai updates