Copilot CLI /fleet speeds app porting benchmark

GitHub’s terminal agent stack is maturing into a real orchestration layer, but the benchmark shows speed gains are only “cheap” when the task graph is actually parallel.

• –Autopilot enables multi-step execution without constant user turn-taking, which is useful for long migration-style tasks.
• –`/fleet` can cut wall-clock time by dispatching independent subtasks to parallel subagents.
• –Parallel subagents can increase LLM interactions, so premium-request usage may rise versus simpler loop-based flows.
• –For tightly coupled, sequential app ports, loop-based control can remain competitive on cost and reliability.
• –The practical playbook is to benchmark by task shape, not hype: parallelizable refactors favor `/fleet`, dependency-heavy ports may not.