Why In direct FSK the keying signal shifts the transmitter's oscillator (VFO) itself, so the radiated carrier jumps between the mark and space frequencies with no audio stage involved. Audio FSK instead feeds two audio tones into an SSB transmitter's microphone or line input, and the SSB mixer translates those tones up to RF. The final on-air signal has the same shift, the same bandwidth and the same baud rate either way; the difference is only where in the chain the frequency shift is generated. Direct FSK avoids audio-stage distortion and overdrive that can create AFSK sidebands, but it is not inherently narrower or faster.
Watch out The bandwidth and baud rate choices are tempting because AFSK can sound dirty when overdriven, but occupied bandwidth depends on the shift and keying speed, not on how the shift is produced, so those claims are false and the all-of-these choice fails with them.
Direct FSK shifts the VFO; AFSK shifts audio tones into an SSB rig. Same bandwidth, same speed, different injection point.