A clean way to interpret the univariate OLS slope, Cov(X,Y)/Var(X), is "out of all the variation in X (/Var(X)), how much of that is related to Y (Cov(X,Y))?"
Logic extends to multivariate (X'X)^(-1)X'Y if you remember that inverse = divides by = "out of"