Once you are confident with building ML pipelines, you will realize that there are many mundane routines that you have to perform to prepare features and tuning hyperparameters. You also will feel more confident with certain methods, and you will have a pretty good idea of what the techniques are that would work well together with different parameter settings.
In between different projects, you gain more experience by performing multiple experiments to evaluate your processing and modeling pipelines, optimizing the whole workflow in an iterative fashion. Managing this whole process can quickly get very ugly if you are not organized from the beginning.
Necessity of AutoML arises out of these difficult situations, when you are dealing with many moving parts and a great number of parameters. These are the situations where AutoML can help you focus on the design and implementation details in a structured manner.
