At the 2018(?) ICFP, I sat between John Wiegley and Conal Elliot. They talked about expressing and solving a programming problem in category theory, and then mapping the solution into whatever programming language their employer was using. From what they said, they were having great success producing efficient and effective solutions following this process.

I decided to look for other cases where this process worked.

I found several, but one off the top of my head is high dimensional analysis, where t-SNE was doing okay, and a group decided to start with CT and try to build something better, and produced UMAP, which is much better.

In short, this does work, and you can find much better solutions this way.

(random link https://stats.stackexchange.com/questions/402668/intuitive-e... )