It's interesting to think about what the process will look like when we do understand them. I imagine pulling bits of LLM off the shelf like libraries and compiling them together into a functioning "brain", precisely tailored to your needs.