Why do we try to define $dx$ as a linear functional from the tangent space to the real numbers when we use a similar notation for the local coords on the tangent space?
Example, let $M$ be a smooth manifold and $T_pM$ be its tangent space at $p$. Then the linear functional $dx: T_pM \to R$ maps $t \in T_pM$ to $t^i$, $dx^i(t) = t^i$. But when describing the basis vectors on $T_pM$ we write the local coordinates as $\left ( d/dx_i\right)$, so we use $dx_i$ in the bottom too.