Two keys: one fills the square, the other drives the column shuffle.
ADFGVX is two ciphers in a row, which is what made it so strong in 1918.
The 26 letters and 10 digits are written into a 6×6 square, in an order set by the first keyword. The rows and columns are labelled A, D, F, G, V and X, and every character becomes the pair of labels for its row and column. If T sits in row D, column G, it becomes DG.
That string of label letters is written in rows under the second keyword and read off column by column in the keyword's alphabetical order, exactly as in a columnar transposition. The two halves of each pair usually end up far apart.
The labels were chosen because their Morse codes are very different from one another, which cut down on transmission errors. The earlier ADFGX version, from March 1918, used a 5×5 square of letters only, with I and J sharing.
Germany introduced the cipher days before its 1918 spring offensive. French cryptanalyst Georges Painvin broke it by hand in June 1918, working himself to exhaustion. The break revealed where the next attack would fall and helped the Allies stop it. The decrypted message became known as the Radiogram of Victory.
You need both keywords. The ciphertext gives itself away: nothing but A, D, F, G, V and X (or, for ADFGX, just A, D, F, G and X), always an even number of letters in total.
They were chosen because they sound nothing alike in Morse code, so a radio operator would rarely mistake one for another. Each plaintext character becomes a pair of them.
ADFGX, introduced in March 1918, uses a 5×5 square of letters. A few months later the Germans added V to make a 6×6 square, which also holds the digits 0 to 9.
The first fills the square: its letters go in first, then the rest of the alphabet (and digits). The second is a columnar transposition keyword that scrambles the order of the coordinate letters.