In Walter Rudin's Principles of Mathematical Analysis (3rd edition) (page 10), it is proved that
for every $x>0$ and every integer $n>0$ there is one and only one positive real $y$ such that $y^n=x$. (This is number $y$ is then written $\sqrt[n]{x}$.)
In particular, this implies the existence of $\sqrt{2}$.
On the other hand, if one considers the polynomial $f(x)=x^2-2$ as an element in the ring $\mathbf{Q}[x]$, one can adjoin a root of $f$ to $\mathbf{Q}$. The procedure (see, for instance, Michael Artin's Algebra (2nd edition) page 456) is to form the quotient ring $K = \mathbf{Q}[x]/(f)$ of the polynomial ring $\mathbf{Q}[x]$. This construction yields a ring $K$ and a homomorphism $F\to K$, such that the residue $\overline{x}$ of $x$ satisfies the relation $f(\overline{x})=0$.
In the real analysis case, $\sqrt{2}$ can be approximated (or defined, depending on how one constructs the real numbers) by a Cauchy sequence of rational numbers: $$ 1, 1.4, 1.41, 1.414, 1.4142, \cdots $$
In the abstract algebra case, the set of real numbers is absent; one does not even need to define it. And there is no way to "approximate" $\overline{x}$.
These two ways to define the object $\sqrt{2}$ seems to be somewhat different in that the defined object has rather different properties.
How should one understand the "discrepancy" here? Are there other relations/connections between these two definitions besides being a root of $f(x)=x^2-2$?