I am intensively testing ai step. I get pretty nice results, as my understanding of how to split information between goal, tasks, and context is increasing.

However, once the ai step is a bit more complex than just asking for simple information, it happens a lot that it gets stuck. By stuck, I mean that in messenger, the conversation is still open because it does not go to next step, but it does not react anymore.

What intrigue me is that in the playground, it never happens. It can ends the conversation too soon because of bad design in the context / tasks, but that is not what happens as it would go to next step in that case.

Does someone know why it gets stuck and/or knows how to prevent that? And why it does not get stuck in the playground?