Someone pointed out to me today a list of academic papers on feature recognition dating from 1988 ( It is interesting to consider the early work done with geometric models and other directions mainstream modeling might have gone if more interest was paid to these ideas back then. It seems that most of this work with feature recognition was in the context of automating the creation of or feeding parameterized construction history trees which was the popular trend in 3D design at that time. Secondly feature recognition was widely seen as a benefit for generating parameterized machining operation lists.
To me a central part of Kubotek’s novel idea that has sparked the CAD industry’s recent trend toward geometry-based modeling is that the largest benefit of feature recognition is not in the automation of design. Like other projects from the late 80s and early 90s, Kubotek’s software can find patterns of geometric faces and in appropriate cases create parameterized features from them. What is different is that Kubotek software knows that one view of the features is only useful for some of the users of the geometry. Whether original construction steps or recognized from geometry, features are not the central definition of the model, they are always disposable.
This outlook on features prevents Kubotek software from locking designs down to just one user’s view of the features. This translates into valuable design freedom and agility. There certainly are occasions in which design can be locked down and automated but more often difficult problems require constant adaptation to insights gained as the project progresses. Besides does locked down and automated sound like any fun to you?

Source: CAD Freedom and Precision – Views and News from KubotekUSA – Recognizing Features
Go to Source: CAD Freedom and Precision – Views and News from KubotekUSA