Lightning Talk: Program Complexity and Thermodynamics - Vadim Alexandrov - CppCon 2023 

Talk of entropy has a high likelihood of turning into "woo", but this is easily avoided by starting from the definition at 1:22. Start by considering the number of "ways" that something could occur. In this case, we might consider the number of ways something could be implemented, or the number of ways that an error could occur, or the number of ways that a developer could make a mistake, or maybe something else entirely. If it's fair to assume the ways are equiprobable, then the most likely design/error/issue that emerges is the one with the greatest number of ways of occurring. There are at least two ways to deal with this. An approach that's common in mathematics is to try to find a "universal property" where there is only one possible way that something could be implemented to satisfy some desirable property, and if implementations differ in any way, then at least say that those implementations are fundamentally the same thing. This could be interpreted as developing abstractions so that they satisfy some desireable property, and then eliminate ways that implementation details could prevent one implementation from being swapped with another. The second approach is to embrace the randomness, but engineer the situation so that the favorable outcome is the one that has the most possible ways of occurring. I'll have to leave it to the reader to interpret how this could be done, but maybe it could be considered a weaker form of the first approach?