Live streaming platforms like Twitch.tv count millions of viewers and performers, pointing to novel practices and interactions. Instructional applications of this phenomenon are rising with thematic channels and shows aimed at teaching and debating a variety of subjects ranging from art to programming. However, few research publications have specifically explored trends in pedagogical strategies and learning behaviours of viewing audiences. This study responded to this need by analysing computer science shows on Twitch.tv. A total of 40 hours of live streaming were analysed within a Community of Inquiry framework using the concepts of social affordances, computational thinking, and teaching style. Data collected from both expert and novice streamers included streamers' actions, on-screen activity, and online discussions. Results pointed to specific tactics employed by both performers and viewers for exploring computer science issues and dynamics together, with instances of peer-tutoring, construed learning, and significant transparency, even when top-down instruction was staged.