We use it in production on some pretty heavy platforms, and we're considering moving more platforms onto it.
My feedback:
* Read/write performance is as good or better than you will get on any other MySQL platform, and much better in some cases. We didn't really choose Aurora explicitly for it's performance gains so YMMV.
* Replicas had some serious lag issues on earlier versions of Aurora, but they've completely disappeared for us after the 1.6 patch (released in March 2016).
* Peace of mind with replicas is worth a lot. They shouldn't ever fall out of sync with the writer node, they lag less than traditional async MySQL replication (as of the 1.6 patch), & failover to them is pretty fast/easy.
* We had mild stability issues for a while, but those have also largely gone with the 1.6 patch.
* Pitfall: if you're heavy on big temp-tables for on replica nodes, then you might not get great performance on creation & reads from those temp-tables.
* Protip: if you're in a Java-based stack, use the MariaDB JDBC driver (in Aurora mode). This eases farming out queries to replicas, and makes failover a lot easier/quicker.
I'd also say that Aurora isn't a fix for a bad data model and/or lack of indexes. We've noticed that CPU% is pretty comparible to any other flavor of MySQL for heavy/complex query workloads.
I recently spoke about Aurora at an AWS Summit. Video here: https://www.youtube.com/watch?v=qvMKKDbzaxo