A simpler approach that doesn't involve having to "ride the gain" is to operate the rig with AGC switched*on* (despite what the WSJT-X user guide states) and to set the audio level coming out of the rig and going into the ADC to give a level somewhere near the top of the scale on the WSJT-X 'thermometer' when receiving only band noise.
While I agree that use of AGC reduces the need to ride gain, it does NOT eliminate it in the presence of very strong signals. I use an Elecraft K3 and use slow AGC to minimize its effect on decoding. Slow is plenty fast enough.
Rob Sherwood NC0B, he of Rx testing fame, has demonstrated conclusively that the WSJT-X decoder (specifically the WSPR one) can decode weak signals without degradation of their SNRs in the presence of an 'interfering' in-passband signal that's 75dB to 80dB stronger than the wanted signals. I've also verified that weak FT8 signals can be successfully and correctly decoded in the presence of an in-passband interfering signal that many 10's of dB stronger than the wanted signals.
Retired from a long career in pro audio, I didn't need to make any measurements to predict that. It's the nature of A/D conversion. :)
