Calculations in the latent space (medium.com)

Why is the author referring to the most hidden layer in a vanilla autoencoder as the latent space? I thought only variational autoencoder uses the latent space for generative purposes.