Agentic Routing: The Harness-Native Data Flywheel

2026-07-13

AI Summary

This work addresses the limitations of traditional routing methods, which optimize only a static cost–accuracy trade-off and fail to adapt to dynamic execution states, intermediate failures, or feedback loops in agent workflows. The authors propose a dynamic routing mechanism natively integrated into the execution framework that, at each step, selects the optimal single model or model ensemble based on the full execution context to balance cost efficiency and accuracy. Central to this approach is a “framework-native data flywheel,” which automatically converts structured execution trajectories from every routing decision into training data, enabling continuous refinement of both routing policies and specialized models. Leveraging a four-layer routing stack, a LightGBM cold-start ranker, and a staged training pipeline, the system iteratively improves performance using logged arena records. Evaluations on benchmarks such as DRACO and PinchBench demonstrate that the method not only achieves effective cost control but also functions as a native data engine for agent training.

Abstract

Large language model agents are increasingly executed not by a single model call, but by an execution harness that manages observation, context, control, action, state, and verification. At the same time, frontier and open models are becoming structurally specialized: a model that is strong at code editing, long-context recovery, tool use, mathematical reasoning, or low-latency response may not dominate on the other axes. This makes model selection inside an agent a core systems problem rather than a per-query serving trick. Existing routing methods mostly optimize single-turn cost-quality trade-offs and therefore miss the execution state, intermediate failures, and feedback loops that make agents different from chat completion. We propose Harness-Native agentic routing, a step-level routing paradigm that selects either a single best-fit model for cost-effective execution or multiple complementary models for ensemble-style accuracy improvement, conditioned on the full harness state. The key insight is that every routing decision naturally produces a structured data record -- consisting of the query, harness state, model choice or model set, execution trace, outcome, and cost -- whose labels are supplied by the environment rather than by the router itself. These records form a harness-native data flywheel: execution traces train better routers and harness-native models, which improve cost-quality trade-offs and generate more traces under the same budget. We instantiate this idea in OpenSquilla with a four-layer routing stack, an open LightGBM cold-start ranker, and a staged router-model path that turns logged arena records into progressively stronger routing policies. The report studies singleton and multi-model routing on agentic benchmarks including DRACO and PinchBench, and argues that agentic routing is not merely cost control, but a data engine for agent-native training.

Problem

• agentic routing
• model selection
• execution harness
• data flywheel
• multi-model ensemble

Innovation

• Agentic Routing
• Harness-Native
• Data Flywheel
• Step-level Routing
• Model Ensemble