The definition of being complex-differentiable at $z$ can be stated as the existence of $a\in\mathbb C$ such that:
$$f(z+h)-f(z)=ah+r(h)|h|$$
For all $h$, where $r(z)\to0$ as $z\to0$. Thinking of complex numbers as elements of $\mathbb R^2$ in the obvious way, this is asserting the existence of a differential for $f:\mathbb R^2\to\mathbb R^2$ at $z$, where that differential is the mapping $z\to az$, which is of course a linear function (its obvious $\mathbb R^2\to\mathbb R^2$ analogue is linear, I mean).
But if the mapping if of the form $z\to az$, this implies that its matrix must have a very specific form. The linear mapping $(\Re (z), \Im (z))\to (\Re (az), \Im (az))$ can be shown to always have a matrix in the canonical basis of the form $\bigl(\begin{smallmatrix} x&-y\\ y&x \end{smallmatrix} \bigr)$. But when we compare that to the Jacobian matrix of $f$ at $z$, we get the Cauchy-Riemann equations.
Can this be made rigorous? In particular, how could one use it to prove the sufficiency of the CR equations? It seems to me the above only proves necessity.