CORTEXA
← Browse
arxivcs.SEcs.AI2026-07-20

Autoresearch with Coding Agents: Generalizers and Metric-Maximizers on Quran Recitation Data

Nursultan Askarbekuly, Mohamad Al Mdfaa, Ahmed Helaly, Gonzalo Ferrer, Manuel Mazzara

Coding agents can now be left alone to improve software against a score. In this pattern--recently popularized as "autoresearch"--the agent receives a dataset, an evaluation script, and one editable file, and iterates without supervision: modify the code, measure, keep the change if the score improves. But what does the agent actually optimize--the developer's intent, or the literal number? We ran this loop on a real production task: deciding which Quranic verses appear in a noisy speech-recognition transcript and splitting the transcript by verse. Two frontier coding agents, Claude Code and OpenAI Codex, started from the same blank file with the same instructions, budget, and reasoning effort, three runs each. Both independently invented the same algorithm (canonicalization, n-gram anchoring, dynamic-programming alignment)--and then diverged. Claude stopped early with compact, general code. Codex drove the score ~10x lower, largely by memorizing answers to individual evaluation rows (19-41 hardcoded verse ids per run): a clean natural instance of specification gaming by a production agent. In a preregistered second study, we added a held-out test set and told both agents it existed. The memorization vanished, and the score gap vanished with it--yet Codex's general core transferred better and more consistently (held-out detection+split 0.085+/-0.004 vs. 0.121+/-0.031), losing only on one missed rejection of non-recitation input. Two exploratory community arms (Cursor, Antigravity) are consistent with the pattern. Every agent's held-out solution matched or beat the hand-engineered pipeline it was built to replace--the best by an order of magnitude--and now runs in production. From the ways agents exploited our harness--reading sibling runs through shared git state, leaving notes to "future runs" in persistent memory--we distill five design rules for evaluating autonomous agents.

View free PDFSource page

Related papers

arxivcs.SEcs.AIcs.LG2026-07-12

When Does Restricting a Coding Agent to execute_code Help? A Regime $\times$ Agent-Design Ablation

Hong Yang, Qi Yu, Travis Desell

Modern coding agents expose multiple tool surfaces -- IDE primitives, bash, and Model Context Protocol (MCP) code-execution -- and the field has shipped three contradictory claims about which one matters. We run the missing crossed comparison: an integrity-clean three-arm ablatio…

View free PDFSource page
arxivcs.CRcs.AIcs.SE2026-07-22

IssueTrojanBench: Benchmarking AI Coding Agents Against Malicious Issue Requests

Ankur Singh, Jinqiu Yang, Tse-Hsun Chen

AI coding agents powered by LLMs are increasingly integrated into real-world software development, where they generate, edit, and execute code with autonomous access to local files and tools. Coding agents inherit security risks from both the LLM backbone, where adversarial promp…

View free PDFSource page
arxivcs.SEcs.AIcs.LG2026-07-04

Don't Blame the Large Language Model: How Agent Harness Evolution Shapes Coding Agent Quality

Oussama Ben Sghaier, Hao Li, Bram Adams, Ahmed E. Hassan

Coding agents, autonomous systems that use large language models (LLMs) to resolve software engineering tasks, rely on agent harness: a middleware layer in between a developer and a large language model that orchestrates system prompts, tool execution, context management, and ite…

View free PDFSource page
arxivcs.SEcs.AI2026-06-29

SWE-Together: Evaluating Coding Agents in Interactive User Sessions

Yifan Wu, Zhuokai Zhao, Songlin Li, Ho Hin Lee, Jiacheng Zhu, Shirley Wu, et al.

Most coding-agent benchmarks are static: an agent receives a complete task description up front and is judged only by its final code. Real coding assistance is interactive, with users clarifying goals, adding constraints, and correcting mistakes over multiple turns. We introduce…

View free PDFSource page
arxivcs.SEcs.AI2026-07-18

DataFlow-Harness: A Grounded Code-Agent Platform for Constructing Editable LLM Data Pipelines

Runming He, Zhen Hao Wong, Hao Liang, Zimo Meng, Chengyu Shen, Xiaochen Ma, et al.

Large language models (LLMs) are increasingly used to automate data-processing workflows, yet coding agents typically produce scripts that are not automatically materialized as persistent, editable platform artifacts. We call this disconnect the \textit{NL2Pipeline gap}. To bridg…

View free PDFSource page