Model WorkFlow Prompt Engineering

In the second video of the course Andrew Ng mentions that prompt engineering going to slowly replace the ML workflow like for eg; Considering a supervised Learning model it requires the following steps : 1. Get label Data 2.Train Model on Data 3.Deploy and call the model the entire process would take 3-4 months collaborating between various teams however with prompt engineering Andrew mentions that simply by calling the prompt engineering one can call the model api and get the desired output and that could be done in a matter of hours , question is the model still needs to be trained and deployed as it is done traditionally so how will prompt engineering will save time ?

I think the takeaway is that people won’t be acquiring and cleansing their own data or training their own models. Merely tuning how they extract desired responses from existing LLMs.