I don't think you are making the point you think you are making, or you being unclear. Take the concurrency example. The is a big difference in understanding the high-level concept of parallelism vs serial execution, which you suggest any library user understand, vs. understanding the C++ techniques and raw code that interacts with low-level threads to provide the library functionality. A good library does not require its users to understand its implementation in order to use its API -- that would be ridiculous most of the time. The burden falls on library authors to provide a good API abstraction that does not require users to understand the implementation; though some bad APIs may in fact require this of the users due to poor design.