CrabOS is a research proposal for an operating system architecture designed around the idea that humans and AI agents co-inhabit the same computational environment — sharing state, context, and work in progress rather than treating agents as external callers into a system. The core insight is that current OS abstractions were designed for human users alone, and grafting agents on top creates persistent friction: dropped context, invisible side effects, and handoffs that lose provenance. CrabOS attempts to make shared work state a first-class primitive.
For interaction designers, the most relevant contribution is the framing of context not as something an agent fetches at task start but as a continuous, jointly maintained resource — closer to a shared whiteboard than a prompt. This has direct implications for how we design context panels, memory UIs, and the handoff moment when an agent returns work to a human. The architecture also surfaces questions about agent identity at the OS level: which actions were human-initiated, which were agent-initiated, and how that provenance is preserved across a session.
The evidence base is currently case studies rather than controlled evaluation, so the claims should be treated as design hypotheses rather than validated findings. Still, the conceptual vocabulary CrabOS introduces — co-inhabitation, shared work state, OS-level provenance — is genuinely useful scaffolding for teams working on agentic product architecture and MCP-style infrastructure.