It has been an interesting experience working on FSet and this book over the last eight months or
so, while LLM-driven programming has exploded onto the scene. As I promised on the title page, this
book contains no LLM-generated text whatsoever. I wasn’t going to use an LLM to generate any code
for FSet either, but at one point, curiosity got the better of me and I had Claude Code try to write
one function (ch-bag-tree-less). I checked it meticulously; it was very close to correct,
but not perfect. At that point, I was almost done with the major coding anyway, so I went back to
doing it by hand.
Programming manually has come to feel a little quixotic, but I’m still glad I did it. It’s clear that LLMs don’t yet have the design sense of a skilled human. Writing this book, I came across a number of points where I thought improving the code was a better choice than documenting its current behavior, so I made those improvements; I don’t think an LLM, at least in early 2026, would have had the judgment to know when to do that, and if I weren’t writing the book myself, I might not have noticed the issues in the first place.
At the very least, FSet can be my small contribution to the world’s human-generated training data.
FSet is now faster, better tested and debugged1, more feature-complete, and with this book, much better documented than it has been for the bulk of its nearly-two-decade history.
I think FSet makes writing Common Lisp easier and more enjoyable. I hope it not only finds wider use within the Common Lisp community, but also helps grow that community by attracting people who have not yet discovered the pleasures of the language.
— Scott L. Burson, April 2026