[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-study-finds-ai-code-agents-stumble-on-rewritten-code":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},5746,"study-finds-ai-code-agents-stumble-on-rewritten-code","Study Finds AI Code Agents Stumble on Rewritten Code","A new benchmark shows top AI coding agents lose accuracy on rewritten but equivalent code, and no model is robust across every tool.","Rename a few variables and rewrite some loops, and today's best AI coding agents get measurably worse at fixing real bugs.\n\nResearchers tested two agent scaffolds, mini-SWE agent and OpenCode, each running four frontier models: Claude Opus 4.5, Kimi K2.5, MiniMax M2.5, and Qwen 3.6-27B, on issues from SWE-bench Verified and SWE-bench Pro. They built a sampler that rewrites the surrounding codebase into a logically identical but differently structured version, swapping identifier names, adding dead code, and restructuring control flow, then ran each agent on both the original and rewritten version of the same task. Comparing paired results isolated the effect of the rewrite from ordinary run-to-run noise. The worst-hit setups saw resolve rates drop by as much as 6.7 percentage points, with 6 of 16 model-scaffold-dataset combinations showing a statistically significant decline.\n\nThe bigger finding is not the drop itself but its unpredictability. Qwen 3.6-27B was among the sturdiest models under mini-SWE agent but the most fragile under OpenCode, so no model can be called simply more robust than another; the ranking flips depending on the harness around it. Real codebases already vary this much in structure and style without anyone trying to trip up the agent, so this is the kind of variation these tools will meet in ordinary use, not some contrived adversarial test.\n\nThe simpler of the two scaffolds, mini-SWE agent, held up best overall, a reminder that a leaner harness can matter as much as the model running inside it.","[\"ai-agents\",\"code-generation\",\"llm-evaluation\",\"software-testing\"]","2026-08-20T04:00:00.000Z","2026-08-20T05:24:37.249Z","2026-08-20T05:24:49.182Z","published",null,[],"ai",[26,27,28,29],"ai-agents","code-generation","llm-evaluation","software-testing",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.18389",0,{"sections":36},[37,40,44,49,54,59,64,69,74,79,84,89,94,99],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",3293,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",435,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",210,"2026-08-19T09:32:27.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",140,"2026-08-19T18:25:42.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",95,"2026-08-18T16:05:00.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Science","science",90,"2026-08-19T18:41:02.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",73,"2026-08-18T07:51:50.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":78},"Dev Tools","dev-tools",69,"2026-08-18T04:00:00.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",33,"2026-08-18T22:18:13.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]