I think it's the other way round: If someone like Knuth can deal with levels of complexity others can't then he should work on highly complex problems. Others are perfectly capable of writing tools that "do one thing well".