We moved from structured decoding to tool calls and agents a long time ago. It was more useful when models were too weak to hold syntax reliably.

Yeah. Myself, a PhD in math, have less attention to detail than models.

Now their main weakness is creativity, planning and decision making

Techniques in DSPy are still extremely valuable. Your comment is akin to saying, we don't need feedback because open loop control is so good now. It is until it isn't. I don't need a deep understanding of the underlying theory because my gut reaction is always right, until it isn't.