An agent harness is top-down: one model, one operator, delegated tasks. An agency harness is networked and horizontal: it coordinates from the edge, and agency emerges between harnesses, not inside the model.
Today's agentic stacks give software harnesses: coding harnesses, research harnesses, memory harnesses. All of them are agent harnesses in structure. An operator delegates a task, the model executes it, the result returns to the operator. Whatever the tooling, the topology is a hub: one model, one owner, capability accumulating at the center.
The economic harness is the other kind. It does not exist to execute delegated tasks; it exists to make its holder an economic agent — to define value, form commitments, create economic spaces, access liquidity, share surplus. Its function is to distribute the power to make economic state-changes, not to concentrate the power to act on someone's behalf.
The contrast runs on four axes. Whose goals: the operator's, versus the edge's own. Where coordination lives: inside one model's context, versus between harnesses, through protocol. What the model is: the protagonist, versus a shared semantic field, a translation device between agents' local meanings. What accumulates: the capability of a central model, versus the agency of a network.
The name marks the difference deliberately. An agent harness makes software act for you. An agency harness makes you, whether person, network, or AI agent, an economic actor in your own right. The register discipline travels with the distinction: the model is never "intelligent," agency is produced through use rather than granted, and no token confers it. The user's mental model: LLM plus harness equals a view into the network as a whole, with users and agents at the endpoints.
Related: · · ·