Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AdaptOrch: Task-Adaptive Multi-Agent Orchestration in the Era of LLM Performance Convergence

Geunbin Yu

TL;DR

AdaptOrch is presented, a formal framework for task-adaptive multi-agent orchestration that dynamically selects among four canonical topologies based on task dependency graphs and empirically derived domain characteristics and establishes orchestration design as a first-class optimization target independent of model scaling.

Abstract

As large language models from diverse providers converge toward comparable benchmark performance, the traditional paradigm of selecting a single best model per task yields diminishing returns. We argue that orchestration topology -- the structural composition of how multiple agents are coordinated, parallelized, and synthesized -- now dominates system-level performance over individual model capability. We present AdaptOrch, a formal framework for task-adaptive multi-agent orchestration that dynamically selects among four canonical topologies (parallel, sequential, hierarchical, and hybrid) based on task dependency graphs and empirically derived domain characteristics. Our framework introduces three key contributions: (1) a Performance Convergence Scaling Law, formalizing conditions under which orchestration selection outweighs model selection; (2) a Topology Routing Algorithm that maps task decomposition DAGs to optimal orchestration patterns in O(|V| + |E|) time; and (3) an Adaptive Synthesis Protocol with provable termination guarantees and heuristic consistency scoring for parallel agent outputs. We validate AdaptOrch across coding (SWE-bench), reasoning (GPQA), and retrieval-augmented generation tasks, demonstrating that topology-aware orchestration achieves 12-23% improvement over static single-topology baselines, even when using identical underlying models. Our results establish orchestration design as a first-class optimization target independent of model scaling.

AdaptOrch: Task-Adaptive Multi-Agent Orchestration in the Era of LLM Performance Convergence

TL;DR

AdaptOrch is presented, a formal framework for task-adaptive multi-agent orchestration that dynamically selects among four canonical topologies based on task dependency graphs and empirically derived domain characteristics and establishes orchestration design as a first-class optimization target independent of model scaling.

Abstract

As large language models from diverse providers converge toward comparable benchmark performance, the traditional paradigm of selecting a single best model per task yields diminishing returns. We argue that orchestration topology -- the structural composition of how multiple agents are coordinated, parallelized, and synthesized -- now dominates system-level performance over individual model capability. We present AdaptOrch, a formal framework for task-adaptive multi-agent orchestration that dynamically selects among four canonical topologies (parallel, sequential, hierarchical, and hybrid) based on task dependency graphs and empirically derived domain characteristics. Our framework introduces three key contributions: (1) a Performance Convergence Scaling Law, formalizing conditions under which orchestration selection outweighs model selection; (2) a Topology Routing Algorithm that maps task decomposition DAGs to optimal orchestration patterns in O(|V| + |E|) time; and (3) an Adaptive Synthesis Protocol with provable termination guarantees and heuristic consistency scoring for parallel agent outputs. We validate AdaptOrch across coding (SWE-bench), reasoning (GPQA), and retrieval-augmented generation tasks, demonstrating that topology-aware orchestration achieves 12-23% improvement over static single-topology baselines, even when using identical underlying models. Our results establish orchestration design as a first-class optimization target independent of model scaling.
Paper Structure (42 sections, 3 theorems, 16 equations, 12 figures, 6 tables, 2 algorithms)

This paper contains 42 sections, 3 theorems, 16 equations, 12 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM Performance Convergence
  4. Static Orchestration Frameworks
  5. Dynamic Model Composition
  6. Parallel Agent Execution in Practice
  7. Concurrent and Recent Work
  8. Problem Formalization
  9. Model Convergence
  10. Task Dependency Graphs
  11. Orchestration Topologies
  12. Performance Convergence Scaling Law
  13. The AdaptOrch Framework
  14. Phase 1: Task Decomposition
  15. Phase 2: DAG Construction
  16. ...and 27 more sections

Key Result

Proposition 1

Let $\mathcal{M}$ be $\epsilon$-convergent on task distribution $\mathcal{D}$. Let $\text{Var}_M$ denote performance variance from model selection and $\text{Var}_\tau$ denote performance variance from topology selection. For a task $T$ with dependency DAG $G_T$ having $k$ subtasks, under uniform su When $\epsilon \to 0$ (perfect convergence) and $\omega(G_T) > 1$ (parallelizable tasks), $\text{Va

Figures (12)

  • Figure 1: Paradigm shift from model selection (left) to orchestration design (right). When model capabilities converge, the dominant optimization variable becomes the structural topology of agent coordination.
  • Figure 2: AdaptOrch pipeline. The Topology Router (Algorithm \ref{['alg:routing']}) selects the optimal execution topology based on DAG structural properties ($\omega$, $\delta$, $\gamma$). Failed syntheses trigger re-routing with adjusted coupling estimates.
  • Figure 3: $\epsilon$-Convergence evidence across four benchmarks. All five models score within $\epsilon$ of the best, validating the convergence assumption (Definition \ref{['def:convergence']}). Dashed line: best model score; shaded band: $\epsilon$ range.
  • Figure 4: Main results comparison across three benchmarks. AdaptOrch achieves the highest accuracy on all tasks while maintaining competitive latency. Error bars show $\pm 1$ standard deviation over 3 runs.
  • Figure 5: Pareto front: accuracy vs. latency. AdaptOrch achieves the best accuracy-latency tradeoff across benchmarks, dominating other methods in the Pareto sense.
  • ...and 7 more figures

Theorems & Definitions (10)

  • Definition 1: $\epsilon$-Convergence
  • Definition 2: Task Dependency DAG
  • Definition 3: DAG Structural Properties
  • Definition 4: Canonical Topologies
  • Proposition 1: Orchestration Dominance under Convergence
  • proof : Proof sketch
  • Corollary 1
  • Definition 5: Consistency Score (Heuristic)
  • Proposition 2: Synthesis Termination
  • proof