The downfall was fragmentation. They merged Keras after TF started having multiple competing application libraries, each managed by different people (and each f

The downfall was fragmentation. They merged Keras after TF started having multiple competing application libraries, each managed by different people (and each fighting for promotion, as you say). Keras just happened to be the most popular. Even after Keras, various teams (e.g., deepmind) have decided to make their own libraries.