The reason it's possible to move fast here is due to the fact that I'm porting an existing language to a new runtime. All the hard work and hammock time has already been spent by the core Clojure team and the community. I'm not aiming to reinvent the wheel here. Furthermore, having worked with Clojure for nearly two decades now, I have a very good idea of what the language does, and what I need to implement on the host platform.

Once I had the basics bootstrapped, it was really a matter of getting tests passing, and then getting more and more libraries to run on it which allowed me to exercise their own tests. And that's what really made the whole thing possible. It's sort of the ideal scenario where you have very well specified inputs and outputs, and a ton of existing tests to ensure you are implementing the behaviors correctly.

The only other aspect here is performance, and that's addressed by having an extensive benchmark harness which compares with JVM timings. This allows me to see where performance gaps are, and focus on addressing them.

I would be careful in extending my experience of being able to port Clojure to Chez Scheme runtime to working with LLMs in general because in most cases you don't have such well defined parameters when you start.