Microsoft Azure Research and the University of Illinois Urbana-Champaign published a paper analyzing 13.5 million coding sessions from 3.2 million users of GitHub Copilot. The study, titled “Agentic Coding in the Wild: Characterizing GitHub Copilot at Production Scale,” draws on one week of production traces from June 2026, during which Copilot generated 761 million LLM calls, 775 million tool invocations, and processed 95 trillion tokens. The median session involves about 15 LLM calls per turn, with the mean exceeding 40 in some scenarios. About 87% of all LLM calls are agent-initiated. KV-cache hits at roughly 90% within a single turn, drops to around 55% after a new turn, and craters to 8% when models are switched or context windows are compacted. The researchers identified five distinct user archetypes, with a 50-fold range in token consumption between lightest and heaviest users. Tool-failure-driven retry loops can amplify compute workload demands by up to four times. The study found that 86-90% of total idle time could be captured by lightweight predictors. The paper suggests scheduling should move from per-request to per-turn or per-session granularity, infrastructure should account for retry amplification, and idle-time prediction should become a first-class infrastructure concern.