The usual presentation of the diagonal argument takes place on the interval $(0,1)$ (we later construct a bijection from $(0,1)$ to $\mathbb{R}$, like $x\mapsto\tan(\pi x-\pi/2)$). The crucial part of the argument is that, given a number $0.a_1a_2a_3...\in(0,1)$ (and all numbers from this interval are of this form), where $a_i\in\{0,1,2,...,9\},\forall i\in\mathbb{N}$, if $f(n)=x_n$, where $f$ is our supposed bijection from $\mathbb{N}$ to $(0,1)$, then we can construct a number $y=0.b_1b_2b_3...$ by making $b_n$ different than $c_{nn}$ (and preferably different than $9$), where $c_{nn}$ is the $n$-th number in the decimal expansion of $x_n=0.c_{n1}c_{n2}c_{n3}...$ In such case, $y\in(0,1)$ and $y\not=f(n)\forall n\in\mathbb{N}$ (because they differ at least in one place), which is the contradiction we wanted. Notice, however, that we needed the assumption that the $n$-th digit always existed.
That is, it makes sense to speak of the $n$-th digit in the decimal expansion of any number in that interval, and it makes sense for such a number to have all digits different than zero (provided they aren't all nines). Natural numbers, however, do not enjoy this same property, since any natural number can be written as a finite sum of ones, even if you were to think of them as having infinite digits (to the left), only a finite amount of them would be different than zero. Therefore, for our "$y$" to be different than every $f(n)$ at least once, it will have "infinite non-zero digits", in which case it's not a natural number. To see this, note that, if $y$ where to have finite non-zero digits, that is, $y=A_kA_{k-1}...A_2A_1;$ $A_i\in\{0,1,2,...,9\}\forall i\land A_k\not=0$, then it wouldn't be different than the number $A_kA_{k-1}...A_2A_1$ which is in the image $f(\mathbb{N})$ by assumption.