Hey @Tr_n_Minh_Quang
The closest thing that exists is resuming. Every subagent execution returns an agent ID, and you can pass that ID back to keep the same subagent going with its context, tools, and model preserved:
Resume agent abc123 and analyze the remaining test failures
That covers the follow-up-and-iterate part of what you’re describing, including after the subagent has finished. It just still goes through the parent chat rather than living as its own session, which is the piece you’re actually asking for. Details in Subagents → Resuming subagents.
We’ll be tracking this post to gauge interest, which helps our product team prioritize. Two open threads are circling the same area if you want to add your voice there too: Agent-spawned side sessions (its V3 section proposes promoting subagents to full sessions) and Having fork of the main agent as subagent.