You want a worker whose findings land back in the conversation you are steering, and you want it to keep going after you close this session. Which part of that is available, and why?

The principle — where a worker reports and how long it lasts are the same fact seen twice, so asking for both of these is asking for two different surfaces at once.

Why the key is correct — the reporting lines are stated together: subagents report results back to the conversation that spawned them, and agent view sessions report only to you. And the lifetime follows from the first: subagents work within a single session. A worker that reports into this conversation is a delegated worker inside one session that does a side task in its own context and returns a summary — which is precisely what cannot outlive the session it belongs to. The wish is not unreasonable, it is just two surfaces.

Why the others are wrong — a subagent cannot span sessions, so the durable half is unavailable that way. A background session is durable but reports to you rather than into the conversation that started it, so the reporting half is unavailable that way. And capability is not the axis at all: in every approach the workers are Claude sessions.

Remember this — decide which you need more, the findings coming home or the work surviving you. You cannot have both from one worker.

Sources — Anthropic's Claude Code agents and subagents documentation.