[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-smarter-way-for-ai-agents-to-tell-web-pages-apart":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},9349,"a-smarter-way-for-ai-agents-to-tell-web-pages-apart","A smarter way for AI agents to tell web pages apart","A new rule for merging duplicate states in AI agent memory judges pages by shared actions, not appearance, lifting task success with no training needed.","A new technique teaches AI agents that two web pages can look identical and still behave completely differently.\n\nResearchers describe an action-conditioned bisimulation rule for the memory systems used by AI agents that navigate graphical interfaces like web pages. The problem: agents that remember pages by how they look often conflate pages that are visually similar but functionally distinct - two tabs in the same widget, or two rows in the same menu, can look alike yet respond differently to the same click. The new rule instead merges two remembered states only when they share the same available actions and those actions lead to matching outcomes, with no training involved and no comparison of visual appearance at all. The team swapped this merge rule into an existing memory system and tested it head-to-head against the original approach on MiniWoB++, a standard benchmark of simulated web tasks.\n\nThe action-conditioned version raised the agent's success rate over a memoryless baseline, while a version without action conditioning, a prior merge rule based on successor representations, and a control that took the same exploratory detours all failed to produce gains. That isolates the fix: it is specifically judging pages by what they do, not what they resemble, that helps. For a field racing to ship browser and GUI agents, this is a reminder that many of the failures live in memory and state representation, not in the underlying model.\n\nIt is a small benchmark, not a browser agent anyone would trust with their inbox yet, but it is a useful data point: agents built to click buttons need memory systems that understand behavior, not just pixels.","[\"ai-agents\",\"gui-agents\",\"agent-memory\",\"benchmarks\"]","2026-10-01T04:00:00.000Z","2026-10-02T11:48:39.119Z","2026-10-02T11:48:42.412Z","published",null,[],"ai",[26,27,28,29],"ai-agents","gui-agents","agent-memory","benchmarks",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.38778",0,{"sections":36},[37,41,46,50,55,60,64,69,74,78,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",5694,"2026-10-01T12:05:27.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",821,"2026-10-01T14:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":45},"Policy","policy",432,{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",311,"2026-10-01T14:19:00.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",197,"2026-10-01T11:37:06.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":18},"Science","science",165,{"name":65,"slug":66,"count":67,"latest_published_at":68},"Consumer Tech","consumer-tech",150,"2026-10-01T11:59:27.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":75,"slug":76,"count":72,"latest_published_at":77},"Software","software","2026-09-30T21:41:11.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":45},"Startups","startups",86,{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]