Length for AES256 symmetric encryption passphrase?

[Barnet Wagman]

> You'd want a key derivative function that produce an output of 32
> bytes to use as the actual AES key. But you are indeed correct in the
> point that what matter is the amount of entropy provided by the
> passphrase.
>
>
How long a passphrase is recommended for generating a 32 byte (AES) key? 
   I'll probably generate the passkey programmatically, so typing an 
human memory are not issues.