Anyone have an opinion on https://code.claude.com/docs/en/agent-teams How do we think OpenHands sdk lines up?

I'm curious how much of this is the model vs. agent facilitating things like "Teammates communicating directly with Teammates". For example, could we reliability re-create Agent Teams with non-Anthropic models?

Some context on this question: https://openhands-ai.slack.com/archives/C06R25BT5B2/p1771260728322749?thread_ts=1771257467.436299&cid=C06R25BT5B2 TL;DR: Anthropic is leapfrogging OpenAI here. It re-created

mail + beads

, which works wonders with GPTs, in a hardcoded workflow for what they call 'tasks' and 'subagents' for Claudes. They even credited

beads

this time. Seems likely the latest Claude has some training on this workflow.

I have a few research projects at CMU on agent delegation, and I think that it works with other models too, although probably would need some tuning