Killer @claude_code session just now. Agents ran for 55 minutes straight just now. Had a detailed plan to update a bunch of stuff, and then had it think hard about sequencing subagents to work. No compacting either! Here's a mid-session screenshot ~50 minutes in…

Claude used 8 subagents, one of them 2x. "code-reviewer" performed final checks before completion. Zero compacting the whole time. How? Subagents "compress" context. Each has its own context window to work, and only has to report back a summary to the main agent.