While I’m by no means an experienced coder, a close friend recently introduced me to @DSPyOSS and RLMs, and I'm absolutely going to spend some time this year learning both. The whole approach is completely Adam Smith pilled and it's such an intuitive illustration of how complex systems achieve efficiency. Specialization makes workers faster at individual tasks but more importantly the separation of concerns allows each stage or component of a project to be optimized independently. Combining DSPy and RLMs does exactly that: instead of summarizing bulk context (which is inefficient and leads to information loss), RLMs pro-actively delegate context to Python scripts and sub-LLMs. This is comparative advantage applied to inference! Central planners in the mud once again.