Real-Time AI Service Economy: A Framework for Agentic Computing Across the Continuum

A research team including Lauri Lovén, Schahram Dustdar, and five other authors has published a framework addressing a critical challenge in the emerging AI agent economy: how to efficiently allocate computing resources when autonomous AI services compete across device, edge, and cloud infrastructure. Their paper, 'Real-Time AI Service Economy: A Framework for Agentic Computing Across the Continuum,' reveals that the structure of service-dependency graphs—modeled as directed acyclic graphs (DAGs)—is the primary determinant of whether decentralized, price-based markets can function reliably at scale. When dependencies are hierarchical (tree or series-parallel structures), prices converge to stable equilibria and optimal allocations can be computed efficiently. However, when dependencies become complex with cross-cutting ties between pipeline stages, prices oscillate and allocation quality degrades, making systems difficult to manage.

To solve this instability, the researchers propose a hybrid management architecture where 'cross-domain integrators' encapsulate complex sub-graphs into resource slices that present simpler, well-structured interfaces to the broader market. Their systematic ablation study across six experiments (1,620 total runs) demonstrates that this hybrid approach reduces price volatility by 70-75% without sacrificing throughput. The research confirms that under truthful bidding, decentralized markets can match the allocation quality of centralized value-optimal baselines, provided appropriate mechanism design with quasilinear utilities and discrete slice items is implemented. This work provides both theoretical foundations and practical architecture for building scalable, efficient markets where AI agents can autonomously request and pay for compute resources in real-time across distributed systems.