It has been argued that we increasingly live in a regime of anticipation in which likelihoods and probabilistic outcomes prevail. Many settings, ranging across finance, social media, biomedical science and military planning, rely on a semi-automated form of statistics – sometimes called ‘machine learning’ – to generate the predictions on which anticipation relies. Anticipation takes hold as these settings incorporate predictivity focused on the attributes of populations and individuals. What kinds of subjects live in the regime of anticipation? Shifts in predictive practice directly index the re-shaping of subjectivity in anticipation. Exploring the use of machine learning in social media, this article examines predictive practice of anticipation. It shows how software developers and programmers not only become agents of anticipation, but also internalise regimes of anticipation through technical practices. Shifts in programming practice hint at what it is like to be an agent of anticipation.