[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-more-thinking-beats-more-tools-for-ai-coding-agents":10,"sections":34},{"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":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3546,"more-thinking-beats-more-tools-for-ai-coding-agents","More thinking beats more tools for AI coding agents","A 90-run study finds that raising reasoning effort nearly triples first-try success rates, while a browser testing tool added cost but no accuracy gains.","Giving AI coding agents more tools does not make them more reliable — spending more on reasoning does.\n\nResearchers ran 90 independent agent sessions, each tasked with building the same real-time retrospective board from a single detailed spec. Every run was scored against a fixed 14-criterion rubric worth 42 points. The study varied model generation, agent harness, reasoning effort, access to a browser-based testing tool, and design-oriented system prompts. Frontier models clustered near the top of the rubric; a low-cost local model scored between 24 and 37 points. The single most common failure — container deployment — fell apart on the first try in 44 percent of runs, and its failure rate tracked model generation more closely than it tracked any other variable. Raising reasoning effort from High to xHigh lifted first-try perfect runs from 28 percent to 89 percent and cut corrective follow-up prompts roughly fivefold, at 9 to 29 percent additional cost. The browser testing tool, by contrast, raised cost 42 to 68 percent while moving functional scores by nothing. A design-focused system prompt lifted visual quality scores from 3.0 to 4.5 on a 5-point scale — but a one-paragraph paraphrase of that prompt reproduced the full gain, suggesting the lift came from explicit direction, not prompt engineering craft.\n\nThe result lands awkwardly for vendors selling agentic coding products on the basis of tool integrations and scaffolding complexity. If most first-run failures trace to weak reasoning rather than to flaws a checker could catch, the ROI case for elaborate toolchains weakens considerably. Compute spent on deeper reasoning appears to outperform compute spent on verification loops.\n\nThe study is observational and limited to one task type, so generalizing too broadly would be a mistake — but the directional finding is hard to ignore: before adding capabilities, ask whether the model is actually thinking hard enough.","[\"ai\",\"dev-tools\",\"research\",\"agents\"]","2026-07-03T04:00:00.000Z","2026-07-03T08:23:59.966Z","2026-07-03T08:24:02.865Z","published",null,[],"ai",[24,26,27,28],"dev-tools","research","agents",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02436",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":26,"count":77,"latest_published_at":78},"Dev Tools",59,"2026-07-07T04:00:00.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]