For the past decade or so, the role of the data science teams has been relatively straightforward - collect and refine data, work on creating and training the models, build proof-of-concepts then hand it over to the application developers. Lately, with the advent of the Transformer models and with the LLMs in particular, building an NLP functionality into a new or a legacy product has become a necessity and a must-have. How is the application development lifecycle changing, is there an 'agile NLP,' what new tasks suddenly appeared, and what is the role of a data team in this fast-paced world where models have just become a commodity?