To my understanding it's to be considered as an extra, similar to SSE.
AVX works with SSE or non-SSE code, so in that sense it is backward compatible. It added a lot of other features in addition to SSE e.g. wider vector, 3 operand instruction coding, compact and efficient encoding of instructions, all adding to give better performance.
Thank you both for your responses!
One more thing to add: There is a performance penalty for switching from SSE to AVX instructions (and vice-versa), so you should not mix them too much in your code. Try grouping them in separate blocks or use AVX only where possible.