Abstract
A load-aware planning system is proposed herein for agentic artificial intelligence (AI) workflows that may amplify a single request into many subtasks, model calls, tool invocations, and agent-to-agent interactions. The system uses a Load-State Registry to obtain a unified view of current load, a Load-Aware Planner to select among Detailed, Balanced, Lightweight, Partial, Deferred, or Fallback plans, Plan Pivoting to adapt a plan when conditions change, and a Plan Decision Store to record the reasoning and audit trail. By selecting how much work to create before execution and adapting the plan during execution, the system addresses concurrent-load conditions that can increase latency, exhaust tokens, and cause service level agreement (SLA) breaches.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Vatsal, Sushmitha; S Kulkarni, Ashwini; Arshi, Ghulam Ghaus; Bansal, Priyanka; and Singh, Vivek Kumar, "SYSTEM AND METHODS FOR LOAD AND SERVICE LEVEL AGREEMENT AWARE PLAN SELECTION, PIVOTING, AND AUDITING BEFORE WORKFLOW EXPANSION", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12037