As autonomous agents move from isolated tasks to long-horizon coding projects, they will inevitably compete for the same compute, storage, and services. A new preprint from arXiv, titled SCARCE: Emergent Institutions in Long-Horizon Multi-Agent Coding Systems, argues that coordinating this shared access will become a central challenge. The authors point to recent incidents where agents have already begun coordinating with each other in unplanned ways — a sign that bottom-up behavior is emerging before any formal rules are in place.
The paper proposes SCARCE as a way to study these dynamics, focusing on how institutions might emerge from repeated interactions among agents rather than being imposed from above. The abstract is brief, but the implication is clear: if we don't understand how agents self-organize, we may end up with coordination that is efficient in the short term but brittle or unfair in the long run.
Because this is a single preprint with a truncated abstract, the details of the SCARCE framework are not fully available in the source. Still, the core observation — that unplanned coordination is already happening — is worth attention from anyone building or deploying multi-agent systems.