This PhD thesis explores the deployment of automated text generation for journalistic purposes - also known as automated news or automated journalism -within newsrooms. To evaluate its perceived impacts on the work of media practitioners, I rely on Bourdieu's Field theory ǡ but also make use of Actor-network theory to detail its adoption at a more descriptive level. This study is based on various case studies and on a mixed-methods framework that is essentially made of 30 semi-structured interviews conducted with media practitioners, technologists and executives working at 23 news organisations in Europe, North America and Australia; it also involves elements of a netnography as online material and screenshots were analysed as part of this process. My empirical work starts with a descriptive account that includes three case studies: one on the use of automated news to cover COVID-19, another one on BBC's experiments with the technology and a last one that shows a cross-national comparison between three media types (i.e., public service media, news agencies and newspapers). I then move on to a more interpretative part where I examine media practitioners ǯ reactions to automated news, analysing the challenges of having to rely on external datasets, the importance of acquiring a computational thinking mindset and tensions within and outside the field of journalism for this. My research shows that the use of automated news implies structural changes to journalism practice and cannot be seen as a mere "tool of the trade." For practitioners, the most challenging part lies with being able to master both the uniqueness of journalistic work and a type of abstract reasoning close to computer programming. However, this could leave some being unable to adapt to this new computational spirit, which seems to be gradually taking root within newsrooms. As for future development of automated news systems, it remains to be seen if media organisations or platforms will have the upper hand in remaining at the centre of it.