I will answer you question at the end. Please go through this first to get a context of my answer:
RSA is an asymmetric encryption algorithm. So, it generates a private/public key pair.
When you say you are sharing A's keys to B, you must only share A's public key. Similarly, B's public key must be shared with A. This can be done by putting the public key in a certificate and sending the certificate. Or, public key can also be sent in a text file as it is. But A's private key will remain with A and B's private key will remain with B.
Now, what A needs to do is generate a random number say X, encrypt X with B's public key and send it over to B. B will decrypt it with its private key.
B will generate a random number Y, encrypt Y with public key of A and send it over to A. A will decrypt it with its private key.
At this point, both A and B will have X ad Y. Now, they can generate an AES session key out of X and Y(say as simple as by multiplying them and forming XY). This session key can be used to encrypt/decrypt application data.
Usually, this is how an AES session key is generated. RSA provides a way to exchange the components of a key and the components are put together to form the real key. Real key is used in AES for encryption/decryption.
Now coming to your question, you're asking "is it acceptable for both A and B to be pre-configured with a secret AES-256 key (instead of secret RSA-2048 keys) that is used to generate an AES-256 session key?"
The answer is, the whole point of using RSA is to act as a key exchange mechanism to derive an AES key. If both A and B are pre-configured with the same AES key already, there is no point in using RSA. You can directly encrypt/decrypt using the AES key. Anyway, AES is faster than RSA(http://forums.windowsecurity.com/viewtopic.php?p=81177 and symmetric encryption session keys in SSL/TLS).