When do message queues make an architecture too complicated?

Hey everyone, I've been thinking about message queues lately, stuff like SQS, RabbitMQ, or Kafka. People always say they're the best way to separate services, deal with background jobs, and handle big traffic bursts. And yeah, a lot of the time that's true. But I'm starting to wonder, when do they actually make things *harder* to manage and more complicated than they need to be? For a mid-sized microservices app, maybe with a few async tasks but not huge scale, when do you think a message queue is just too much? Are there certain project sizes or team setups where running a queue system is more trouble than it's worth? What have you all found, keeping things simple versus bringing in a queue early on?