Like any automated tool, it's a matter of control, and whether you like the decisions the algorithm makes on your behalf.
Occasionally, you have very particular needs for a UV unwrap that the algorithm might not be optimized to serve.
For instance, I once had to set up a flowing lava river for a Wii game. We achieved the fluid flow effect by scrolling the textures along the mesh, which meant that I needed to ensure that "downstream" always aligned with "positive v" in the UV space. This involved contorting the winding river in very strange ways within the UV texture space, and going well outside the normal 0...1 bounds (since the lava textures tiled) - stuff that an automatic unwrapper wouldn't think to do out of the box, and it might be difficult or impossible to instruct it to do so. We also had to apply small aesthetic touch-ups here and there to hide stretching in inconspicuous parts of the mesh, or speed up & slow down flow based on the level art, which necessitated going in by hand.
More common cases where this might arise could be when you're hand-painting textures for your models in a conventional 2D texture map, rather than a 3D painting program or procedural material generation system. In a situation like that, it's helpful if the model's UVs are laid out in a way that's intuitive to the texture artist - similar parts grouped together, oriented in a sensible way - again stuff that an automated unwrapper might not know about or prioritize in the quest for efficient packing and uniform texel density.
Like any hypothetical, it's hard to give a concrete answer about when you should or should not use an automatic unwrap. It's easier to work on concrete examples case-by-case. If for your current task, the unwraps you're getting from the automated tools seem to be serving your needs, great! If it ain't broke, don't fix it. If you run into a case where you're not getting what you want from the automated tools - for any idiosyncratic reason that might arise in your particular project - then it will be clear that you need a different strategy.
So, don't go trying to solve that problem until / unless it actually becomes a problem for your work. ;)