Get lost in Machine Learning

I have been doing the machine learning project and exploring the kernel from kaggle. I just found that the more I have read, the less confidence I have. There are soooo many models are available and so many methods for feature engineering. Just got lost on how to choose and what to do. For example, since XGboosting can handle the missing values well, do we still need to impute? Since lesso can determine the important features, is it still necessary to exploring the data? the recent auto ML can fit different model itself and pick the best one, then what data scientist need to do?
There are also so many new tec and new improvement keep coming up, neural network, tensorflow ,keras… IT tech develops so fast, I found it is difficulty to catch-up and easy to get confused as a new learner. What is the right strategy for ML? Which is the right direction to pursue the DS career ? I am really appreciated if someone can provide guidance and advice. Thank you!