This is a Plain English Papers summary of a research paper called CoworkOS: A Persistent Runtime for Delegating Long-Horizon Agent Work. If you like these kinds of analyses, you can find more AI and machine-learning research on AIModels.fyi or follow us on Twitter.
CoworkOS targets work that must both deliver and persist
CoworkOS separates task strategy from execution management: a model chooses actions, delegation, and acceptance, while a runtime preserves state and feedback. Its central technical claim is that a three-layer Main Agent, Coordinator Runtime, and Subagents can support complex artifact production alongside continued monitoring, rather than treating long-running work as a sequence of disconnected chats.
The system is designed for tasks such as producing software, presentations, CAD models, or Blender scenes while also tracking processes and decisions over time. Humans can join through conversation or by directly editing the same evolving artifact. CoworkOS retains working files, decisions, and Job logs across agents and sessions, so an interrupted task can resume from recorded state.
The overview links complex productivity and continued monitoring to shared work, benchmark comparisons, and recursive self-improvement. These components summarize CoworkOS’s combined long-horizon and artifact-delivery goals....
