GitHub found compressing tool output made Copilot tasks more expensive.

GitHub published four measured cost reductions on September 2: selective output compaction 5.5%, removing view-tool line-number prefixes 3.1%, a compacted task-tool prompt 2.9%, batching background-task results 2.3%. The counterintuitive result came from RTK (Rust Token Killer), which shortens shell output: individual responses got smaller and tasks got more expensive, because the agent reopened the original output or reran the command to recover what was cut. Their conclusion is directly reusable: "tokens per tool call is the wrong objective," because cost has to be measured over the whole task.