Claude Code now supports background sub agents that continue running even when you press escape, allowing you to interact with the main conversation while multiple agents work in parallel. This feature, available with the Sonnet 1M token context model via API, enables spawning 10+ sub agents that work autonomously and report back when completed.
Background sub agents represent a new capability in Claude Code where spawned agents continue working even after the user interacts with the main conversation. Previously, spawning sub agents meant waiting until all tasks completed before you could do anything else. Now, the agents run as true background tasks.
The feature appears to be tied to the Sonnet 1M token model accessed via API. When you give Claude a large task requiring multiple parallel operations, it can spawn sub agents that:
The creator demonstrates by asking Claude to spin up 10 sub agents to create SEO pages. The agents work in parallel, researching topics and generating content. As each agent completes, it reports back, and the orchestrating agent can process results incrementally - even triggering additional actions like browser testing via MCP.
A side benefit of the 1M token model is the elimination of constant context compacting. The creator notes going from 2% context remaining on the standard model to having virtually unlimited headroom, removing the need to run "slash compact" commands.
"This is a whole new thing that I've never seen before. I don't know if this is just the Sonnet 1 mil or if this is a new feature that they've kind of shipped without telling people."
The power comes with a price tag. Running extensive sub agent workflows with the 1M token model can burn through tokens quickly:
The creator suggests that being able to provide direct input to individual sub agents while they're running would make the system even more powerful - turning it into the "most OP system ever."
"So now background tasks. We can actually see the 10 sub agents working in the background, right? That's so cool, guys."
"You no longer have to worry about escape cancelling everything, right?"
"Previously, you'd have to wait an hour, wait for every single sub agent to finish, and then you could have a conversation with Claude. That is no longer the case."
"This is a great update from Anthropic. They're cooking something over there. Don't know what they're cooking, but it smells good in the kitchen."