How to do dimension reduction from a variational autoencoder

I am thinking about a variational autoencoder. As far as I understand it, in the encoding section you compress to a px1 tensor and then you create a $\mu$ and $\sigma$ of dimensions of my choice (though less than $p$ ). The decoding layer is then the reverse. But what if I want to do dimensionality reduction using the VAE? Clearly I don't randomly sample from $\mu$ and $\sigma$ . Do I just go to the px1 tensor? Thanks!