My original goal in making the Ext software was to replicate the features present in Bob Bruner's and Amelia Perry's resolvers but to distribute the software in a way that was more flexible and easier to use. In particular the constraints were: 1. For performance reasons, use a "low level" performance-oriented programming language. 2. For research purposes, it should admit high performance use-cases. 3. Most mathematicians don't like to install software so it should have a low-performance web version. 4. For flexibility and ease of experimentation, it needs ergonomic bindings to a dynamic scripting language (I chose Python). 5. Ideally, the web version should also include the scripting language. This turned out to be a rather greedy set of constraints. In the end, trying to make condition (5) work turned into a viable career for me. I will also describe my thoughts on the problem of mathematics software maintenance, and whether language model tools could make ambitious math software projects easier to maintain.