AI coding assistants are transforming software development, yet empirical understanding of how developers integrate these tools into cognitive workflows remains limited. Through computational autoethnography, we analyze 802 collaborative sessions comprising 85,370 messages and 27,672 tool invocations across 47 projects over 30 days. Our analysis reveals several key findings. First, we observe high cognitive delegation, with a delegation score of 0.71 indicating that developers treat AI as a cognitive extension rather than mere autocomplete. Second, we find evidence of intentional model selection, where developers consciously match AI capability to task complexity, resulting in 7.59× longer sessions for high-capability models. Third, sustained collaboration intensity emerges through 2,846 messages per active day and 13.3 projects per week, demonstrating deep integration into development practice. Fourth, developers exhibit context-fluid operation, rapidly switching between projects with minimal cognitive overhead. Fifth, hierarchical tool usage patterns show that execution and exploration dominate, with planning emerging primarily in complex tasks. We introduce the concept of Cyborg Cognition—the integrated cognitive system formed when human direction-setting combines with AI information gathering and execution. This framework extends theories of distributed cognition to human-AI programming collaboration. Limitations include single-subject design; we provide sensitivity analyses and invite replication.