Replacing regressor instead of 2SLS

Hi, can someone explain me why it is usual to run a 2SLS if we are not "satisfied" with one variable and using an instrumental variable to adjust the model? (I am just talking about the just-identified case)

Wouldn't it be easier just to replace the variable with the new one, so we have a new model?

For example:
CO2=beta1 x Energy + beta0

Now I have data of "Energy from non-sustainable sources" which is obviously correlated with "Energy" (here I ignore that "Energy" might be endogenous as well). So, I learned to use "Energy non-sus" as IV and run a 2SLS regression. Intuitively, it would make more sense just to replace the old variable and pretend this variable never existed.
2SLS= going from the wrong model to the right one
Just replacing= having the right model from the beginning

Where is my reasoning wrong?

Best regards