This is a long-winded way of saying that this entire transformation went through code. It's a pipeline that starts with human input, produces code, does an LLM as a judge step and iterates. And you can take this transformation and apply it to a general task as well.

Tools: Code Is All You Need | Armin Ronacher's Thoughts and Writings

This is an unlock I've also personally had. Determining where to use the LLM in your toolchain is going to be a continuous, changing process. However, right now, for me - it's at a place where it can generate code, be used to evaluate a process and then orchestrate within set parameters.

Where this helps me mostly is in the code generation part. Where it's not gotten close to solving problems is thinking better / harder than me. Yet.