There is a particular kind of learner this course was made for, the person who has watched the theory-heavy courses, understands roughly what a convolution or an embedding is, and now just wants to sit down and actually build the thing in a real framework, and for that person this is close to ideal. From the first course you are writing TensorFlow and Keras code rather than watching equations get derived, and the whole specialization is structured around getting your hands dirty on the three workhorse problem types, computer vision, natural language, and sequence and time series data, so by the end you have personally built image classifiers, text models, and forecasting models rather than just read about them. Laurence Moroney is a big reason it lands, because he is friendly, extremely clear, and good at keeping the cognitive load manageable, introducing one new idea at a time so you rarely feel buried. It also maps neatly onto the TensorFlow Developer certification exam if having that credential matters to you.
The flip side of being this practical is that it is deliberately light on theory, and that is the thing to be clear-eyed about. You learn how to call the APIs and assemble working models, but it will not teach you the maths of why they work or take you deep into architecture design, so on its own it can leave you able to build without fully understanding, which is a shaky place to stop. It stays at the applied level throughout, and because TensorFlow and Keras keep evolving, a little of the material lags the newest versions, so expect to reconcile the occasional API change. My recommendation is to treat this as one half of a pairing rather than a complete education.
Do a concept-first course such as Andrew Ng's for the understanding, do this for the fluency, and the combination leaves you genuinely able to both reason about models and build them, which is exactly where you want to be.