Natural-Language Agent Harnesses
2026-03-26 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors look at how the part of an agent that controls actions, called the harness, is usually mixed into code and hard to separate or study. They propose writing the harness instructions in plain natural language, making it easier to edit and share, calling this approach Natural-Language Agent Harnesses (NLAHs). They also create a system called Intelligent Harness Runtime (IHR) to run these natural language harnesses consistently. The authors test their approach on different tasks to see if it works well and how different parts contribute to performance.
agent harnesscontrol logicnatural language processingruntime systemmodular designbenchmark evaluationcode-to-text migrationagent controllersoftware engineeringexecutable artifacts
Authors
Linyue Pan, Lexiao Zou, Shuo Guo, Jingchen Ni, Hai-Tao Zheng
Abstract
Agent performance increasingly depends on \emph{harness engineering}, yet harness design is usually buried in controller code and runtime-specific conventions, making it hard to transfer, compare, and study as a scientific object. We ask whether the high-level control logic of an agent harness can instead be externalized as a portable executable artifact. We introduce \textbf{Natural-Language Agent Harnesses} (NLAHs), which express harness behavior in editable natural language, and \textbf{Intelligent Harness Runtime} (IHR), a shared runtime that executes these harnesses through explicit contracts, durable artifacts, and lightweight adapters. Across coding and computer-use benchmarks, we conduct controlled evaluations of operational viability, module ablation, and code-to-text harness migration.