Yes and no. It depends. Maybe.
A Gray code, as the article states, is only referenced by the Hamming distance between codes, not the bit size of the encoding. If you repeat the two-bit code: 00,01,11,10 90 times, you'll get a one-degree resolution around a circle, which is plenty for a steering wheel and usually works just as well as a 9-bit encoding (512 states).
The problem becomes "how do I know where the wheel is at any arbitrary time?"
The answer is: you continuously monitor the wheel by counting up or down on each code transition, and have a Home sensor to set the initial position. But now you have to remember the position when you power it off, and perhaps make sure no one can move the wheel while you're not monitoring it. Is that sufficient for your application? (hint: it is for 90% of such applications) If so, problem solved. If not, then we need to get more complex.
More complex can mean: Non-Volatile Memory so we can power off and remember the last position[1], continuous battery-backup so even if main power is off, we track the wheel position, or ta-da: a 9-bit encoding so even with power off, we can encode 512 discrete states. Historically, that last solution was by far the most expensive.
1: I've worked with such a system where the position of a critical gantry was updated while it was running, then when power was off, the last (X,Y) position of the gantry encoders was remembered. Battery-powered electronics then monitored the state of an access door. If the door was opened, then the X,Y position was invalidated (someone could have moved the gantry while the encoders weren't being read) and the software would not allow the system to restart until a manual alignment procedure was done to reset the gantry to a known position and clear the "lockdown" bit. The alternative would have required two 10-bit absolute encoders that cost more than the electronics to do what I described.