for gas density results are very close (within 0.5%) with the advantage for AGA to be faster (up to 10:1),
the main advantage for GERG (2004 or 2008) is that it can model liquid phase (VLE, volume)
there are reports indicating that AGA has problems converging close to a dew point but it
may depend from some specific coding and, with Prode Properties, I have not observed those problems.
When solving complex problems which require many evaluations of gas density AGA should be preferred, in my opinion.
For VLE Prode includes several alternatives to GERG (2008),
for HC dew points if you compare GERG with extended PR results are in most cases within 1-2 K