@keithpjolley I suspect they're wildly different tools with wildly different audiences. In fact, I think that's the root cause of this. By creating systems that don't distinguish between "intended to be interpreted by a machine" and "intended to be read by a human", the moment you start building dependencies on top of readable outputs they're frozen in amber, too risky to change.

Structured output for integration v. prose for humans frees us to improve the human message without that risk.