This paper provides a mathematical formalization of the orchestration state, detailed algorithmic analysis of the execution loop, and controlled benchmarks comparing flat and hierarchical routing under increasing tool catalogs, multi-step workflow pressure, and visible schema-token exposure per LLM call.
Abstract
The rapid expansion of capabilities in Large Language Model (LLM) agents has exposed a critical architectural bottleneck: when agents are given access to a flat, monolithic registry of tools, the model must evaluate hundreds or thousands of options simultaneously. This leads to decision-space explosion, context window saturation, and degraded routing accuracy. To address these limitations, this paper presents a hierarchical, skill-based architecture for agentic orchestration. Capabilities are organized as a rooted tree where internal nodes make routing decisions and leaf nodes execute deterministic tasks. The runtime enforces a single-step execution loop governed by a Last-In-First-Out (LIFO) stack, giving the agent a form of memory akin to a Pushdown Automaton, therefore enabling it to track nested execution contexts and resume deterministically from any depth. Capability discovery follows a manifest-driven, lazy-loading protocol: only the immediate children of the active node are loaded, so memory and prompt costs scale with the explored path rather than the global registry. By replacing global memory with localized stack frames, the architecture prevents outputs from one execution branch from leaking into another, establishing the isolation guarantees required for deployment in regulated enterprise environments. We also discuss UPI Help, an AI-powered digital payments support product, as a motivating production deployment context. We provide a mathematical formalization of the orchestration state, detailed algorithmic analysis of the execution loop, and controlled benchmarks comparing flat and hierarchical routing under increasing tool catalogs, multi-step workflow pressure, and visible schema-token exposure per LLM call.
This survey presents a systematic taxonomy and technical review of dynamic orchestration strategies designed to address communication overhead, KV cache management challenges, and increased token consumption within large Language Model-based Multi-Agent Systems.
Heet Nagoriya, H. Raithatha· International Journal of Kno...· 0 citations
It is shown that LLM-driven agents can violate this condition and introduced a canonical deployment wrapper that guarantees it for arbitrary base agents while preserving already-equivariant behaviour, and it is proved that computing canonical representations required by this construction is graph-isomorphism-hard.
Dynamic agent harnesses let language models change the software that shapes their own execution. This flexibility brings a new reasoning burden: a local plugin change can propagate through dependencies and cleanup. We introduce CordisBench, a 1,200-question benchmark of this lifecycle reasoning. It combines a controlled formal setting with programs executed against Cordis, a runtime that manages component dependencies and cleanup, and asks models to identify affected components, predict state after a specified teardown order, determine which conditions hold under all or some orders, and choose reconfigurations that succeed when executed. Across these tasks, we evaluate three efficiency-oriented models at low reasoning effort with 2, 4, 8, 16, 24, or 32 relevant interactions, using deterministic task-specific scoring. Models usually handle small systems well but grow less reliable as more interactions become relevant, especially when predicting final state and when reasoning across teardown orders. Additional inference effort recovers marked gains for some models. The cost is nontrivial: on our 16-interaction subset, GPT-5.6 Luna uses nearly 3,000 reasoning tokens per question at medium effort. For these controlled instances, that cost is avoidable: an independent finite reference semantics agrees with Cordis execution on every observation and action outcome used for scoring across all 528 executable questions.
FL-MAESTRO is proposed, a multi-agent orchestrator that makes the joint runtime FL decision directly through three specialist LLM agents, one per decision dimension, and matches the accuracy of the strongest energy-aware baseline while cutting wasted round energy from over a third to near zero.
Jiajun Wu, Zirui Wang, Jiayu Zhou et al.· 0 citations
AgentRadio is presented, an asynchronous message-passing layer that equips coding-agent harnesses with three primitives: threads, messages, and waiting for mentions that shows the gain growing with task difficulty, consistent with mid-course correction as the underlying mechanism.
Xinxing Ren, Qianbo Zang, Ziyan Wang et al.· arXiv.org· 0 citations
The more typical feature of agentic AI systems is dynamic, multistep workflows where autonomous components plan, reason, and communicate with external tools and data sources in a series of iterations. Such flexibility increases capability but also brings nondeterminism which is inherent and where the same inputs can result in different execution paths and outputs. The variability creates a major challenge to the traditional observability approaches that are mostly created to support deterministic and service-oriented architectures. This paper redefines observability as an architectural element and introduces a trace-oriented architecture to suit agentic processes. The suggested solution presents semantically rich trace units capturing reasoning transitions, the intent to select a tool, the evolution of memory, and policy interactions, to gain a better insight into the execution behaviour. One of the major innovations is the combination of branch-aware trace modelling with an adaptive fidelity mechanism to dynamically change monitoring granularity based on uncertainty and anomaly indicators. Experimental analysis of various agent workflows shows significant increases in completeness of traces, accuracy in detection of anomalies and localization of root-cause, and a significant decrease in diagnosis time. The findings suggest that the suggested architecture does not only increase interpretability but also helps to ensure the reliable and efficient functioning of non-deterministic AI systems.
Ankur Gupta, Karan Gupta, Divyakumar Deepak Savla et al.· International Conference on...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
