As large language models grow more capable, AI agents are shifting away from isolated, domain-specific tasks toward long-horizon workflows that span multiple domains. According to a new paper posted to Hugging Face, that shift exposes two problems: harness complexity is rising faster than manual design can scale, and tight coupling to a single domain limits how general any one harness can be.
The paper reframes the central question — instead of asking how to engineer a stronger harness for one domain, it asks how to autonomously construct specialized harnesses, improve them through experience, and orchestrate them across domains.
Its answer is Raven, described as "The Harness of Harnesses," an open-source multi-agent ecosystem that automatically constructs and evolves modular harnesses for specific models and domains. The paper treats each executable model–harness pair as a composable unit of intelligence.
To support what the authors call an All-Domain Collaboration Network, Raven's Host Agent decomposes goals, matches subtasks to specialized agents, coordinates execution dependencies, and integrates results. A host archive and EverOS preserve experience across tasks, while Skill Forge turns that experience into reusable procedures.
The paper also presents a theory establishing sufficient conditions under which such composition expands reliable task coverage beyond that of the available individual agents under a shared resource budget. On complex, long-horizon tasks, the authors report that Raven significantly outperforms state-of-the-art agent systems.