In subspaces of $U \subseteq \mathbb{R}^n$ the goal is to define the tangent space in a geometric sense as a space that is orthogonal to the space $U$ at a point $p \in U$. This can be visualized in cases such as $S^2 \subseteq \mathbb{R}^3$, where the tangent space at a point $p \in S^2$ is just the space of points that are orthogonal to $p$, that is, $$O_p=\{x \in \mathbb{R}^3 \ | \ \langle x,p \rangle =0\}.$$ I am not sure if this is the way it's done, but I guess a standard way to generalize this could be to define the tangent space of a subspace $U \subseteq \mathbb{R}^n$ at $p \in U$ to be the set $\{x \in \mathbb{R}^n \ \vert \ \langle x,p \rangle=0\}.$ When trying to generalize the spaces we are working with, one doesn't have the nice structure of a subspace of $\mathbb{R}^3$, which is why one needs to adapt definitions. One way that can be done is by working with smooth functions as the following standard way does. Let $M$ be a smooth manifold. Then the tangent space at $p \in M$ is $$T_pM=\{ v:C^{\infty}(M) \to \mathbb{R} \ \vert \ v \ \text{linear and satisfies the product rule}\}.$$
I wonder, in what sense the intuitive geometric view and the abstract manifold definition are compatible. What one probably wants is for these two definitions to be isomorphic, such that they share the same properties as vector spaces and we have a way to identify them by an isomorphism. In the case of the sphere this is quickly done: the sphere is a $2$-dimensional manifold and it is well known that the tangent space $T_pS^2$ is then also $2$ dimensional. Since $O_p$ is $2$-dimensional as well, we get an isomorphism $O_p \cong T_pS^2$. Can this be done for all subspaces of $\mathbb{R}^n$ and is a linear isomorphism suffienct to capture the properties that we want to capture?