At the beginning of AR, it was necessary to place special markers in the scene. They looked like a simpler version of a QR code. This was the only possibility to reconstruct stable 3D positions over multiple frames.
With better computer vision algorithms and better camera quality, the specialized markers were replaced by arbitrary images.
And finally, markerless tracking was invented. It means, that virtual content can be placed anywhere in the scene. No special marker, image, or target is necessary.
The simplest version of markerless tracking is based on plane detection. More complex algorithms create a full 3D reconstruction of the environment. Users can place virtual objects anywhere on the reconstructed model.
Nowadays, every AR framework offers markerless tracking.
When is it useful?
AR apps with markerless tracking do not need any additional props. No images or objects. Users can just start the experience in any environment.
This is especially useful when the virtual content is not tied to a specific object in the real world. When the real world just serves as a stage. For example, if you want to examine a 3D model of a house. The 3D model just needs a stable pose.
Are markers still a thing?
Marker-based tracking is useful, whenever a specific real-world object triggers the augmented reality experience. This can be a book page, a magazine cover, or a toy packaging. But also a full 3D model such as a car. Even small QR-like codes can be useful, e.g. if you want to augment a wide range of products with additional information.