Leveraging LLM-Based Agentic Systems to Generate Quantum Applications for Test Optimization

Jul 16, 2026

Quantum computing is increasingly applied to software engineering optimization, yet converting natural-language requirements into executable quantum applications still requires deep quantum and programming expertise. QPipe introduces an LLM-based multiagent pipeline that autonomously transforms natural-language requirements into traceable quantum-application workflows via specialized agents spanning parsing, formulation, code generation, review, execution, and verification. Evaluations on 20 requirements show 100% compilation and 96.7% successful execution with final-result combination, averaging 260.1 seconds and 1.89M tokens per requirement. Successful applications often outperform a genetic-algorithm baseline, and ablations confirm the importance of code-generation capability, task knowledge, review feedback, and multiagent decomposition.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

Introduction

Background

QPipe

Evaluation

Related Work

Conclusion

arXiv :2607 .00939v 1 [ cs . SE] 1 Jul 2026 Leveraging LLM-Based Agentic Systems to Generate Quantum Applications for Test Optimization Ming Tao 1 , Yuechen Li 1,* , Tao Yue 1,* , Man Zhang 1 , and Aitor Arrieta Marcos2 1 Beihang University, {taoming, liyuechen, yuetao, [manzhang}@buaa. edu. cn](manzhang}@buaa. edu. cn) 2 Mondragon University, [aarrieta@mondragon. edu](aarrieta@mondragon. edu) * Corresponding author Abstract Quantum computing is increasingly explored for software engineering (SE) optimization, but translating natural-language (NL) task-level requirements into executable quantum applications still demands substantial quantum and programming expertise. We present QPipe, a large language model (LLM)-based multiagent architecture that autonomously turns NL requirements into traceable quantum-application workflows through specialized agents for requirement parsing, formulation, code generation, review, execution, and verification. We evaluate QPipe on 20 NL requirements, each associated with a real-world benchmark anda test-optimization problem. QPipe successfully completes the key stages of quantum-application generation across requirements, achieving average rates of 100% for code compilation and 96.7% for application execution and final-result combination, with average generation costs of 260.1 seconds and 1.89M tokens per requirement. Among the generated quantum applications that execute successfully, the returned solutions outperform the offline genetic algorithm baseline in most cases. Ablation results further show that QPipe’s advantage depends on retaining code-generation skills, task knowledge, review feedback, and multi-agent decomposition. These results indicate that agentic coordination can support generation of executable quantum applications for tackling test optimization problems from real-world benchmarks. Keywords: Large-language Model, Code Generation, Quantum Computing, Test Optimization, Quantum Workflow 1 Introduction Quantum computing (QC) is increasingly explored for software engineering (SE) optimization problems [40], many of which are classically studied in search-based software engineering (SBSE) [15] . Recent work has applied quantum optimization to SE tasks such as TCO, including test case minimization (TCM) with quantum annealing (QA), which searchs low-energy states of an encoded objective, and test case selection (TCS) with quantum approximate optimization algorithm (QAOA) [28, 27] . These results suggest that quantum optimization is becoming a plausible execution target for SE tasks, but still requires substantial expertise to use effectively. The main challenge is not implementing a quantum routine itself, but translating a natural language (NL) requirement into an executable application with problem formulation, encoding, solver selection, orchestration, etc. Existing works on quantum workflow engineering offer useful abstractions, including workflow modeling in QuantME [31], pattern-based workflow construction [6], and feasibility-oriented design in Q-READY [37] . In parallel, large language models (LLMs) have shown promise in benchmarked quantum code generation [26, 14, 5], QAOA circuit generation [24], and NL-to-quadratic unconstrained binary optimization (QUBO) transformation [38] . However, they do not directly address generating executable quantum applications from NL task-level requirements of SE optimization tasks. This paper presents QPipe, an LLM-based multi-agent architecture for requirement-to-application generation. Given an NL requirement of an SE optimization task, QPipe produces an executable quantum application together with traceable intermediate artifacts. It structures this process through specialized agents for requirement parsing, quantum-suitability analysis, workflow planning, encoding, code generation, review, execution, and verification. We instantiate and evaluate QPipe on TCO through two problem variants, TCS and TCM, using 10 be

FAQs

What problem does QPipe address in quantum-software engineering?

QPipe addresses the difficulty of translating natural-language, task-level requirements into executable quantum applications, including formulation, encoding, solver orchestration, and verification steps.

How does QPipe generate a quantum application from a natural-language requirement?

It uses an LLM-based multiagent architecture with specialized agents for requirement parsing, quantum-suitability analysis, workflow planning, encoding, code generation, review, execution, and verification, producing traceable intermediate artifacts.

What were the main evaluation outcomes for QPipe?

Across 20 requirements, QPipe achieved 100% code compilation and 96.7% for execution plus final-result combination, with average generation cost of 260.1 seconds and 1.89M tokens per requirement; successful solutions often exceeded a genetic-algorithm baseline.

Jul 16, 2026

Upload by Olivia Brown