[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-technique-cuts-compute-costs-for-ai-web-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},5508,"new-technique-cuts-compute-costs-for-ai-web-agents","New Technique Cuts Compute Costs for AI Web Agents","CATTS lets AI web agents add compute only when genuinely unsure, lifting task success up to 11.8% while using fewer tokens than brute-force scaling.","Researchers have found a smarter way to make AI web agents think harder, but only when it's actually worth it.\n\nThe technique, called CATTS (Confidence-Aware Test-Time Scaling), comes from a new arXiv paper studying test-time scaling in multi-step web agents. The researchers found that simply cranking up compute uniformly across every step quickly hits diminishing returns, since small errors compound over long task sequences. They also tested an LLM-based \"Arbiter\" to aggregate multiple attempts, which beat simple majority voting but sometimes overruled decisions the agent was actually confident about. CATTS instead uses the agent's own vote distribution, specifically entropy and the gap between its top two choices, as a signal for when a decision is genuinely contentious and worth extra compute.\n\nThis matters because most test-time scaling approaches treat every step the same, burning tokens on easy decisions and running out of budget on hard ones. Across three benchmarks, WebArena-Lite, Online-Mind2Web, and GoBrowse, CATTS beat majority voting by up to 11.8% while using fewer tokens than naive uniform scaling.\n\nIt is a small paper with a practical idea: agents, like people, should think longest about the calls they are least sure of. Whether it holds up outside these three benchmarks is the open question.","[\"ai\",\"ai-agents\",\"llm-research\",\"test-time-scaling\"]","2026-08-18T04:00:00.000Z","2026-08-18T23:10:17.657Z","2026-08-18T23:10:29.572Z","published",null,[],"ai",[24,26,27,28],"ai-agents","llm-research","test-time-scaling",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.12276",0,{"sections":35},[36,40,44,49,54,59,64,69,74,78,83,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",3293,"2026-08-20T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":39},"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":18},"Dev Tools","dev-tools",69,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",47,"2026-08-19T19:13:46.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",33,"2026-08-18T22:18:13.000Z",{"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"]