Actually, if you have to drive capacitive loads, your amplifier can have a large output impedance and behave mostly as a current source. Such amplifiers are called "Operatioanal Transconductance Amplifiers" (OTAs).
There is nod need for a low impedance output stage. Speed (ft, SR) of OTA is given by the effective transconductance (gm) they represent and the effective capacitance seen. The term effective capacitance stands because in some architectures (Miller like), the dominant pole of your circuit is dominated by the Miller compensation capacitance. In architectures such as symetric or folded cascoded OTAs, the dominant pole is determined by gm and the load capacitance. The gain of OTAs is determined by gm times the effective output impedance (taking into account conductance of output stage and resistive load, if any, attached to the output node). A good picture of OTAs is given by its NOrton equivalent circuit.
To drive resistive loads, your output impedance must be as low as possible. Just think now of your OpAmp as an equivalent Thevenin scheme. A voltage source with an output resistance connected in series must drive another resistor. The resistive divided made by the output resistance of your OpAmp and your load resistor reduce the voltage applied to the latter resistor. That is why your ousput resistance must be lowered by using output stages such as class AB and push-pull architectures. In this case gain and speed of your OpAmp are mostly dependent of internal components (within certain limits). As you must have noticed it, in this case we talk about "Operational Amplifiers" (OpAmps).