I noticed this to be true for nontrivial extensions given by adjoining a square root. (By diagonalizable i mean over an algebraic closure). Given a nontrivial field extension of form $k\hookrightarrow k[\sqrt{\sigma}]$, and an element $\alpha := a+b\sqrt{\sigma}$ of it, the linear operator multiplication by $\alpha$ can be represented, in the basis $(1, \sqrt{\sigma})$, by the matrix $\begin{bmatrix} a & \sigma b\\ b & a\end{bmatrix}$. And, over an algebraic closure of $k$, this matrix can be easily seen to be diagonalizable*.
Is this always the case? Is there any intuitive reason for this to be the case here and/or in general? I'm particularly interested in finite extensions of $\mathbb Q$, but an answer for finite extensions of other fields would be interesting too.
*Actually this can fail in characteristic 2