Coordination primitives for intent-aware orchestration of agent requests

Ascertain whether large-scale, real-time intent-aware orchestration for routing and scheduling AI agent requests requires entirely new coordination primitives or can be achieved through incremental extensions of existing service-mesh technology.

Background

The paper argues that traditional L4/L7 load balancers and service meshes are insufficient for routing based on agent capabilities, jurisdiction, and other constraints. It proposes intent-aware scheduling that matches tasks to agents according to semantic capabilities and policy requirements.

The authors present two possibilities—designing new coordination primitives or extending existing service-mesh technologies—but explicitly state uncertainty about which path will suffice at the required scale and latency.

References

We may discover that entirely new coordination primitives are required, or that incremental extensions of service-mesh technology suffice; at present, the answer is unknowable.

— Upgrade or Switch: Do We Need a New Registry Architecture for the Internet of AI Agents?  (2506.12003 - Raskar et al., 13 Jun 2025) in Unknown Unknowns, subsection “Intent-Aware Orchestration”

Joining the two approaches in a multi-agent LLM orchestrator closing the loop across IoT, edge, and cloud is still untried.

— Smart Adaptive Computing Across the Continuum: LLMs in IoT-Edge-Cloud Resource Management  (2609.09348 - Vaccarella et al., 8 Sep 2026) in Section 5, “Conclusion”

EQ1 --- Decision latency. What is the end-to-end latency of the control-plane decision cycle (intent submission $\to$ decision artifact $\to$ execution dispatch), and how does it scale with the number of candidate capabilities, the complexity of the policy set, and the depth of the context graph? A practical threshold is that control-plane overhead should remain below 10\% of median backend execution latency for the workloads it governs. Micro-benchmarks should isolate candidate generation, policy filtering, and scoring as independent contributors.

— Brain API: An Intent-Aware Control Plane for Policy-Governed Agentic Systems  (2609.21299 - Chernov, 18 Sep 2026) in Section 12.7, Evaluation Criteria (EQ1)