"One FD on" versus both off, or both on
Vilas is correct in asserting that following the OEMs guidance and airline policy is certainly appropriate, and wise, for operating any specific type.
However, in general, and not necessarily related to any specific type, OEM policy, or airline policy, ...there are many circumstances industry wide, where selecting, retaining, or re-selecting "one FD ON" may nonetheless be acceptable, if not even appropriate, and in some circumstances, even necessary. There are specific cases where it is (or was) the policy of an airline, for some aircraft types and circumstances, to select "one FD ON", or to re-select one FD ON. This can be for setting a mode, for protecting mode availability, or for things like presetting the ability to transition to an RNP based missed approach path for a balked landing, in the event of the need for a go-around, or loss of visual reference, or for arming modes or protections, or for retaining some form of FD guidance for a PM, etc.
This issue becomes very difficult for transition between types, or for mixed fleet flying of different types, where habit patterns intervene, and where the philosophy and procedures for FD use may differ, may lead to vulnerabilities, or may outright clash, if not even be mutually exclusive.
So in broad terms, any design where this sensitivity to "both on", "both off", or "one on" becomes an issue, just as Vilas correctly points out, has been known to lead to an accident. So for these designs, is likely that a design improvement would be worth considering, and operating policies and training may need to be reconsidered, to underscore and graphically demonstrate "why" vulnerability exists. Further, such designs may require special care or emphasis during initial or recurrent training, or IOE, to specifically illustrate the risks and vulnerabilities, of not operating according to the OEM's policy or the airline's policy, or per procedures or requirements for that specific type.
Another good example of this principle, of carefully adhering to procedures for a given type, and not assuming that methods from other types will work acceptably, was the previously not emphasized issue of using "rudder checkbacks" and "rudder rolls" the same way in a big jet transport as had been successfully done for decades in tactical fighters, to augment roll control. That pilot information void too, and incorrect assumptions about transfer of procedures from one type to another, once ultimately led to another unexpected fatal difficulty [AA587 - NTSB attributed the accident to the first officer's overuse of rudder controls in response to a wake turbulence event, which caused the vertical stabilizer to fail, and engines to separate from the aircraft.]