Background/Context: While current research provides key insights about successful collaboration in which teachers experience deep learning and practice change, few studies analyze the content of teachers' collaborative conversations about instruction. Even fewer explore how the content of collaborative conversations evolves over time, making it difficult to understand the impact of shifting policies, priorities, and personnel on teachers' collective work. Purpose and Research Questions: To explore teachers' opportunities to learn through collaboration, I draw on the following research questions: How does the depth of teachers' opportunities to learn from collaborative conversations evolve over time? In what ways do contextual factors (e.g., personnel, tools, leadership expectations, coaching) influence these opportunities to learn through collaboration? Participants: This study is situated in two urban middle schools under significant pressure to improve student achievement. I selected one grade-level math team from each school, analyzing their conversations taking place during collaboration meetings. Research Design: To better understand teachers' collaborative learning, I rely on qualitative case study methods. Data collected include interviews and observations of two teacher collaboration groups over four years. I use Horn and colleagues' (2017) taxonomy of teachers' opportunities to learn in conjunction with cultural historical activity theory (Cole & Engeström, 2007) to examine teachers' collective learning over time as this process unfolds in context. Findings/Results: Across both groups, expectations from leadership to improve school performance shaped teachers' opportunities to learn, as did the presence of teachers whose beliefs about math instruction foregrounded student thinking and exploration. Findings demonstrated that for one team, fluctuations in depth of learning also depended on the agendas and protocols used to frame collective work. For the second team, depth of learning opportunities was connected to shifts in personnel and group norms over time. For both teams, high-depth conversations represented no more than a third of the meetings observed in each year, reflecting the complexity of developing effective collaborative cultures supporting deep learning. Conclusions/Recommendations: While high-depth opportunities to learn (OTLs) were infrequent for both teams, the reasons for this scarcity were shaped by each team's context. Leaders hoping to develop collaborative cultures with richer teacher learning opportunities should consider multiple factors shaping OTLs, including collaborative team norms, teachers' beliefs about math instruction, protocols guiding discussions, and the role of accountability pressures in shaping expectations for collective work.