Software is made of building blocks: either an inspired engineer wrote a rules-based engine from scratch, re-used an existing library, or connected and configured the right AWS services.
At some point, those building blocks will become obsolete and in need of replacement. There will always be a better hosting provider, framework, library, or programming language that will turn our codebase into a museum.
How do you know what’s the piece of code that won’t change over time? The most valuable code that you have in your repositories?
First of all, it is industry-specific. It represents a set of instructions that are specific to a field. A K-factor calculation is a good example. The calculation won’t change; it doesn’t depend on the data you have; it stays the same.
It represents a set of rules you use to ingest, analyze, aggregate data. It is specific to your industry, to your company, to the way you do things.
It’s not the ML algorithm that you are using; there are plenty of them. It is the combination of dataset features, how you clean them, and combined together, to produce a specific result that is relevant to your users.
It is company-specific, it doesn’t depend on the data you have, it won’t change over a long time, it is a hardcoded set of rules (not an ML type of try/guest algorithm), and it won’t change even if the technology evolves. That’s your most valuable code.