[{"data":1,"prerenderedAt":3460},["ShallowReactive",2],{"global":3,"home":420,"blog\u002Fself-improvement":1443,"datasource-blog-topics":2247,"engineering-related-articles-ad486cf8-7aee-4c6a-8e7a-c97ac24e846a":2258},{"name":4,"created_at":5,"published_at":6,"updated_at":7,"id":8,"uuid":9,"content":10,"slug":412,"full_slug":412,"sort_by_date":17,"position":413,"tag_list":414,"is_startpage":198,"parent_id":415,"meta_data":17,"group_id":416,"first_published_at":417,"release_id":17,"lang":418,"path":17,"alternates":419,"default_full_slug":17,"translated_slugs":17},"Global","2026-04-14T23:20:08.896Z","2026-08-31T21:01:40.063Z","2026-08-31T22:04:31.034Z",165966679669817,"498ae38d-a57b-4613-a5db-f09eae4c6008",{"_uid":11,"companies":12,"component":155,"demo_form":156,"footer_card":168,"header_card":206,"header_ctas":237,"header_menu":244,"footer_description":270,"footer_link_columns":278,"header_cta_secondary":344,"header_dropdown_menu":351},"20ede2139409",[13,37,51,66,81,96,118,139],{"key":14,"_uid":15,"logo":16,"name":18,"component":21,"hero_title":22,"eyebrow_link":23,"hero_eyebrow":26,"hero_description":27},"finance-tools","b3d51fca-a4c1-48f6-97e2-b9695998d54a",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":20},null,"","asset",{},"company","Trusted agents for the work between your finance systems",{"id":18,"url":18,"linktype":24,"fieldtype":25,"cached_url":18},"story","multilink","For teams already investing in finance automation",{"type":28,"content":29},"doc",[30],{"type":31,"attrs":32,"content":33},"paragraph",{"textAlign":17},[34],{"text":35,"type":36},"Pointer helps finance and ops teams automate the manual steps between ERPs, close tools, procurement platforms, portals, inboxes, and exception queues, with controls and evidence for every run.","text",{"key":38,"_uid":39,"logo":40,"name":18,"component":21,"hero_title":42,"eyebrow_link":43,"hero_eyebrow":18,"hero_description":44},"finance-trust","a98d5e2b-4a5d-4ad7-a711-ea195cc1a5d1",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":41},{},"AI that proves its work.",{"id":18,"url":18,"linktype":24,"fieldtype":25,"cached_url":18},{"type":28,"content":45},[46],{"type":31,"attrs":47,"content":48},{"textAlign":17},[49],{"text":50,"type":36},"Pointer runs the recurring finance and accounting work your team owns, and verifies every result, so you can stand behind what it produces. Purpose-built for teams where being almost right is the same as being wrong. ",{"key":52,"_uid":53,"logo":54,"name":18,"component":21,"hero_title":56,"eyebrow_link":57,"hero_eyebrow":58,"hero_description":59},"finance-savings","ee380e4e-fbc9-424e-90e7-cc0abfa44423",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":55},{},"The finance work that took a team, done without one.",{"id":18,"url":18,"linktype":24,"fieldtype":25,"cached_url":18},"Cut the cost of finance ops, not the throughput",{"type":28,"content":60},[61],{"type":31,"attrs":62,"content":63},{"textAlign":17},[64],{"text":65,"type":36},"Pointer runs reconciliation, three-way matching, and close end to end, so you hit your numbers with a fraction of the headcount. Every run comes with controls and evidence.",{"key":67,"_uid":68,"logo":69,"name":18,"component":21,"hero_title":71,"eyebrow_link":72,"hero_eyebrow":73,"hero_description":74},"finance-accuracy","cb63160e-18dd-4ca3-a94b-723a51b6b827",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":70},{},"AI for the numbers you have to stand behind.",{"id":18,"url":18,"linktype":24,"fieldtype":25,"cached_url":18},"Adopt AI on the work that can't be wrong.",{"type":28,"content":75},[76],{"type":31,"attrs":77,"content":78},{"textAlign":17},[79],{"text":80,"type":36},"Pointer runs the finance work that can't be wrong and proves every figure against your own rules as it happens. You get an evidence trail on every step and nothing to restate at audit.",{"key":82,"_uid":83,"logo":84,"name":18,"component":21,"hero_title":86,"eyebrow_link":87,"hero_eyebrow":88,"hero_description":89},"finance-capacity","e422bd9f-8a98-402f-8fae-a7c348e2dda2",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":85},{},"Serve more clients without adding more people.",{"id":18,"url":18,"linktype":24,"fieldtype":25,"cached_url":18},"Software margins on a people business.",{"type":28,"content":90},[91],{"type":31,"attrs":92,"content":93},{"textAlign":17},[94],{"text":95,"type":36},"Pointer runs the rote prep behind every client deliverable, from reconciliation to NAV to AP across their systems. Each person carries far more clients, and the work still holds up.",{"key":97,"_uid":98,"logo":99,"name":18,"component":21,"hero_title":101,"eyebrow_link":102,"hero_eyebrow":110,"hero_description":111},"computer-use","cv-computer-use",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":100},{},"Your team is the glue between systems. We replace the glue.",{"id":103,"url":18,"linktype":24,"fieldtype":25,"cached_url":104,"story":105},"d76a76c7-7205-48c9-b55a-780f72149194","blog\u002Fsota",{"name":106,"id":107,"uuid":103,"slug":108,"url":104,"full_slug":104,"_stopResolving":109},"A new state of the art for computer use",179031199225579,"sota",true,"Works in any system, with or without an API",{"type":28,"content":112},[113],{"type":31,"attrs":114,"content":115},{"textAlign":17},[116],{"text":117,"type":36},"Your people spend their days copying data between systems that were never built to talk to each other. Pointer uses the screen the same way they do — so it can reach anything they can, and carry the whole process without them.",{"key":119,"_uid":120,"logo":121,"name":18,"component":21,"hero_title":123,"eyebrow_link":124,"hero_eyebrow":131,"hero_description":132},"custom","cv-custom",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":122},{},"Your operation doesn't fit in a box. Stop buying boxes.",{"id":125,"url":18,"linktype":24,"fieldtype":25,"cached_url":126,"story":127},"ad486cf8-7aee-4c6a-8e7a-c97ac24e846a","blog\u002Fself-improvement",{"name":128,"id":129,"uuid":125,"slug":130,"url":126,"full_slug":126,"_stopResolving":109},"Teaching AI systems to learn from their own work",207749086872144,"self-improvement","Built around how you actually work",{"type":28,"content":133},[134],{"type":31,"attrs":135,"content":136},{"textAlign":17},[137],{"text":138,"type":36},"Off-the-shelf tools force you to reshape your workflow to fit their template. Pointer learns how your team actually does the work, encodes it, and runs it your way — so you get automation without giving up what makes your operation yours.",{"key":140,"_uid":141,"logo":142,"name":18,"component":21,"hero_title":144,"eyebrow_link":145,"hero_eyebrow":147,"hero_description":148},"correctness","cv-correctness",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":143},{},"AI that proves its own work — so you can sign off on it.",{"id":125,"url":18,"linktype":24,"fieldtype":25,"cached_url":126,"story":146},{"name":128,"id":129,"uuid":125,"slug":130,"url":126,"full_slug":126,"_stopResolving":109},"Nothing advances unverified",{"type":28,"content":149},[150],{"type":31,"attrs":151,"content":152},{"textAlign":17},[153],{"text":154,"type":36},"Every step Pointer runs is verified against your standard of correct before the next one begins. What can't be proven stops and surfaces to your team. You get the speed of automation with the trust of a manual review.","Config",[157],{"_uid":158,"eyebrow":159,"component":160,"submit_url":161,"placeholder":167},"45be36c0edb3","Request a demo","demo_form",{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":164},"f4aade4d-dea9-4018-b4a1-c5396305013e","contact",{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"Contact",182172365325000,"you@yourcompany.com",[169],{"url":170,"_uid":173,"text":174,"media":191,"eyebrow":204,"component":205},{"id":18,"url":171,"linktype":172,"fieldtype":25,"cached_url":171},"https:\u002F\u002Fpointer.usewrangle.com\u002F","url","99f3ab8cb4a5",{"type":28,"content":175},[176,186],{"type":31,"attrs":177,"content":178},{"textAlign":17},[179,184],{"text":180,"type":36,"marks":181},"Diffuse the frontier.",[182],{"type":183},"bold",{"text":185,"type":36}," We're hiring",{"type":31,"attrs":187,"content":188},{"textAlign":17},[189],{"text":190,"type":36},"across all functions.",[192],{"_uid":193,"type":18,"image":194,"video":199,"lottie":201,"component":203},"38cc782b99df",{"id":195,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":196,"copyright":18,"fieldtype":19,"meta_data":197,"is_external_url":198},166010921746231,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1329x1185\u002Faed12f8247\u002Ffooter-illo-finance.png",{},false,{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":200},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":202},{},"media","Join us","footer-card",[207],{"url":208,"_uid":214,"text":215,"media":228,"eyebrow":204,"component":205},{"id":209,"url":18,"linktype":24,"fieldtype":25,"cached_url":210,"story":211},"fbc7a92e-dc0a-48bf-9a89-8917d6177ae6","careers",{"name":212,"id":213,"uuid":209,"slug":210,"url":210,"full_slug":210,"_stopResolving":109},"Careers",165966648446001,"d8d88108c1d5",{"type":28,"content":216},[217,224],{"type":31,"attrs":218,"content":219},{"textAlign":17},[220,223],{"text":180,"type":36,"marks":221},[222],{"type":183},{"text":185,"type":36},{"type":31,"attrs":225,"content":226},{"textAlign":17},[227],{"text":190,"type":36},[229],{"_uid":230,"image":231,"video":233,"lottie":235,"component":203},"761d663c5ae5",{"id":195,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":196,"copyright":18,"fieldtype":19,"meta_data":232,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":234},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":236},{},[238],{"url":239,"_uid":241,"type":242,"label":159,"component":243},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":240},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"7d507718da1a","DEFAULT_LIGHT","button-link",[245,257,261],{"url":246,"_uid":254,"label":255,"component":256},{"id":247,"url":18,"anchor":248,"linktype":24,"fieldtype":25,"cached_url":249,"story":250},"d39259d6-6adf-4a8f-9d2d-8d9479eb8e31","product","home",{"name":251,"id":252,"uuid":247,"slug":249,"url":253,"full_slug":249,"_stopResolving":109},"Home",165966689307708,"\u002F","e13b01c7-2128-46d9-aa01-dab40f49629e","Product","link",{"url":258,"_uid":260,"label":212,"component":256},{"id":209,"url":18,"linktype":24,"fieldtype":25,"cached_url":210,"story":259},{"name":212,"id":213,"uuid":209,"slug":210,"url":210,"full_slug":210,"_stopResolving":109},"e5ec17427ec3",{"url":262,"_uid":269,"label":266,"component":256},{"id":263,"url":18,"linktype":24,"fieldtype":25,"cached_url":264,"story":265},"f9416751-46cd-4b63-9d45-2e384875fb57","blog\u002F",{"name":266,"id":267,"uuid":263,"slug":268,"url":264,"full_slug":264,"_stopResolving":109},"Blog",165966638812207,"blog","8b56ddef9a42",{"type":28,"attrs":271,"content":272},{"backgroundColor":17},[273],{"type":31,"attrs":274,"content":275},{"textAlign":17},[276],{"text":277,"type":36},"Your finance operation, run by AI that proves its own work.",[279,300,314],{"_uid":280,"links":281,"title":298,"component":299},"a52d884f2f01",[282,286,290,293],{"url":283,"_uid":285,"label":165,"component":256},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":284},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"1575af8c87d8",{"url":287,"_uid":289,"label":266,"component":256},{"id":263,"url":18,"linktype":24,"fieldtype":25,"cached_url":264,"story":288},{"name":266,"id":267,"uuid":263,"slug":268,"url":264,"full_slug":264,"_stopResolving":109},"8f60998b-6b2f-4992-b9bb-f4596c2b497e",{"url":291,"_uid":292,"label":212,"component":256},{"id":18,"url":171,"linktype":172,"fieldtype":25,"cached_url":171},"63f733ab88ca",{"url":294,"_uid":296,"label":297,"component":256},{"id":18,"url":295,"linktype":172,"fieldtype":25,"cached_url":295},"https:\u002F\u002Fstatus.pointer.ai","e113cc65a080","Status","Explore","link-column",{"_uid":301,"links":302,"title":313,"component":299},"fe7afe3fa438",[303,308],{"url":304,"_uid":306,"label":307,"component":256},{"id":18,"url":305,"linktype":172,"fieldtype":25,"cached_url":305},"https:\u002F\u002Fx.com\u002Fpointer","a4df61207b29","X (Twitter)",{"url":309,"_uid":311,"label":312,"component":256},{"id":18,"url":310,"linktype":172,"fieldtype":25,"cached_url":310},"https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fpointer-so\u002F","3b72ade67915","Linkedin","Social",{"_uid":315,"links":316,"title":343,"component":299},"c0a2f9e06735",[317,322,327,335],{"url":318,"_uid":320,"label":321,"component":256},{"url":18,"linktype":24,"fieldtype":25,"cached_url":319},"privacy","ff0ed89c08b6","Privacy policy",{"url":323,"_uid":325,"label":326,"component":256},{"url":18,"linktype":24,"fieldtype":25,"cached_url":324},"terms","1e1f38c03603","Terms of service",{"url":328,"_uid":334,"label":332,"component":256},{"id":329,"url":18,"linktype":24,"fieldtype":25,"cached_url":330,"story":331},"3befeffd-e17f-4958-b5d0-44a6091dc6d4","security",{"name":332,"id":333,"uuid":329,"slug":330,"url":330,"full_slug":330,"_stopResolving":109},"Security",165966720576579,"5c0c3219369f",{"url":336,"_uid":342,"label":340,"component":256},{"id":337,"url":18,"linktype":24,"fieldtype":25,"cached_url":338,"story":339},"2902408f-06ca-477f-86c4-a882eee63ae8","cookies",{"name":340,"id":341,"uuid":337,"slug":338,"url":338,"full_slug":338,"_stopResolving":109},"Cookies",165966658419764,"97398fbed4f0","Legal",[345],{"url":346,"_uid":348,"type":349,"label":350,"component":243},{"id":18,"url":347,"linktype":172,"fieldtype":25,"cached_url":347},"https:\u002F\u002Fplatform.pointer.ai\u002F","345bd7bc7e9d","SECONDARY","Sign in",[352,365,377,389,401],{"_uid":353,"icon":354,"link":358,"component":363,"description":364},"87827214-e723-494e-9483-12fa26e14e99",{"id":355,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":356,"copyright":18,"fieldtype":19,"meta_data":357,"is_external_url":198},166269354530527,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002Fd539b96339\u002Ficon-radial-grey.svg",{},[359],{"url":360,"_uid":362,"label":255,"component":256},{"id":247,"url":18,"anchor":248,"linktype":24,"fieldtype":25,"cached_url":249,"story":361},{"name":251,"id":252,"uuid":247,"slug":249,"url":253,"full_slug":249,"_stopResolving":109},"847c0ffd-4e48-41a0-bf56-b39996af35cc","menu-link","Learn about the Pointer product",{"_uid":366,"icon":367,"link":371,"component":363,"description":376},"1ce7bb0def5d",{"id":368,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":369,"copyright":18,"fieldtype":19,"meta_data":370,"is_external_url":198},166269354587878,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002Fb648e4de8c\u002Ficon-bars-grey.svg",{},[372],{"url":373,"_uid":375,"label":266,"component":256},{"id":263,"url":18,"linktype":24,"fieldtype":25,"cached_url":264,"story":374},{"name":266,"id":267,"uuid":263,"slug":268,"url":264,"full_slug":264,"_stopResolving":109},"c6fb92680fa6","Read our latest thinking",{"_uid":378,"icon":379,"link":383,"component":363,"description":388},"1621c046b828",{"id":380,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":381,"copyright":18,"fieldtype":19,"meta_data":382,"is_external_url":198},166269354583780,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F12c6dbffdb\u002Ficon-explode-grey.svg",{},[384],{"url":385,"_uid":387,"label":212,"component":256},{"id":209,"url":18,"linktype":24,"fieldtype":25,"cached_url":210,"story":386},{"name":212,"id":213,"uuid":209,"slug":210,"url":210,"full_slug":210,"_stopResolving":109},"a2b9999dd5fd","Join our team",{"_uid":390,"icon":391,"link":395,"component":363,"description":400},"569a9894-f1d6-436c-91b1-c5c232679d4d",{"id":392,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":393,"copyright":18,"fieldtype":19,"meta_data":394,"is_external_url":198},166269354542816,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F258d26713f\u002Ficon-corners-grey.svg",{},[396],{"url":397,"_uid":399,"label":165,"component":256},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":398},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"61f1d68a-765c-4f48-b85f-4f3c5c8c7747","Get in touch",{"_uid":402,"icon":403,"link":407,"component":363,"description":411,"hide_on_mobile":109},"dd91bbf7e584",{"id":404,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":405,"copyright":18,"fieldtype":19,"meta_data":406,"is_external_url":198},166269354526430,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F8f7fbab1e3\u002Ficon-up-grey.svg",{},[408],{"url":409,"_uid":410,"label":350,"component":256},{"id":18,"url":347,"linktype":172,"fieldtype":25,"cached_url":347},"0442b6c14028","Log into the Pointer platform","global",-40,[],0,"00375552-dac9-438f-a4d2-01817e029d91","2026-04-14T23:21:10.501Z","default",[],{"name":251,"created_at":421,"published_at":422,"updated_at":423,"id":252,"uuid":247,"content":424,"slug":249,"full_slug":249,"sort_by_date":17,"position":1438,"tag_list":1439,"is_startpage":198,"parent_id":415,"meta_data":17,"group_id":1440,"first_published_at":1441,"release_id":17,"lang":418,"path":253,"alternates":1442,"default_full_slug":17,"translated_slugs":17},"2026-04-14T23:20:11.245Z","2026-08-11T03:51:11.129Z","2026-08-11T03:51:11.149Z",{"seo":425,"_uid":438,"body":439,"component":1437},[426],{"_uid":427,"image":428,"title":432,"og_image":433,"component":436,"description":437},"74d341b84564",{"id":429,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":430,"copyright":18,"fieldtype":19,"meta_data":431,"is_external_url":198},166274089047955,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3600x1890\u002F5551dd8d90\u002Fopengraph-new.png",{},"Pointer",{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":434,"copyright":18,"fieldtype":19,"meta_data":435,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3600x1890\u002Fcc66dbbf8d\u002Fpointer-og.png",{},"Seo","Pointer builds AI systems that do a company’s hardest finance and operations work, and prove it was done correctly – so people can finally come out of the loop.","26260fa6829f",[440,616,644,650,841,934,1010,1097,1099,1160,1235,1276,1373,1376,1402,1404],{"_uid":441,"image":442,"title":445,"entries":446,"eyebrow":588,"variant":18,"component":589,"direction":18,"description":590,"shard_image":596,"eyebrow_link":608,"background_image":610,"shader_config_file":613},"2f2f7480cf2e",{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":443,"copyright":18,"fieldtype":19,"meta_data":444,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F672x380\u002F8947251bac\u002Fhomepage-hero-shader-image.png",{},"Deterministic AI for\u003Cbr>work that must be right.",[447,469,484,499,514,529,543,558,573],{"_uid":448,"image":449,"pointer":453,"component":463,"image_mobile":464,"user_pointer":468},"fbc9c6df310e",{"id":450,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":451,"copyright":18,"fieldtype":19,"meta_data":452,"is_external_url":198},166009497353904,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F942cc8ba64\u002Fhero-product-flow-01-finance.png",{},[454],{"top":455,"_uid":456,"left":457,"message":458,"component":459,"top_mobile":460,"left_mobile":461,"mobile_message":462},"66","4b130cca5e2d","55","Reading email...","pointer-position","59","26","Navigating ERP...","home-hero-slide",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":467,"is_external_url":198},166009497370292,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F792335cda2\u002Fhero-product-flow-mobile-finance.png",{},[],{"_uid":470,"image":471,"pointer":475,"component":463,"image_mobile":481,"user_pointer":483},"ccc91750940a",{"id":472,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":473,"copyright":18,"fieldtype":19,"meta_data":474,"is_external_url":198},166009497366194,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F54f8e8719f\u002Fhero-product-flow-02-finance.png",{},[476],{"top":477,"_uid":478,"left":479,"message":480,"component":459,"top_mobile":477,"left_mobile":477,"mobile_message":462},"50","ae20c4547214","53","Extracting invoice data...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":482,"is_external_url":198},{},[],{"_uid":485,"image":486,"pointer":490,"component":463,"image_mobile":496,"user_pointer":498},"8107840a234f",{"id":487,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":488,"copyright":18,"fieldtype":19,"meta_data":489,"is_external_url":198},166009497403064,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F313e696498\u002Fhero-product-flow-03-finance.png",{},[491],{"top":492,"_uid":493,"left":494,"message":495,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"14","b9ef3f99cfa8","21","Searching for PO...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":497,"is_external_url":198},{},[],{"_uid":500,"image":501,"pointer":505,"component":463,"image_mobile":511,"user_pointer":513},"b09566cef73d",{"id":502,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":503,"copyright":18,"fieldtype":19,"meta_data":504,"is_external_url":198},166009497382582,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F506fa71e6d\u002Fhero-product-flow-04-finance.png",{},[506],{"top":507,"_uid":508,"left":509,"message":510,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"48","4f482b1910b3","11","Reviewing PO-4521...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":512,"is_external_url":198},{},[],{"_uid":515,"image":516,"pointer":520,"component":463,"image_mobile":526,"user_pointer":528},"4acfb167378c",{"id":517,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":518,"copyright":18,"fieldtype":19,"meta_data":519,"is_external_url":198},166009497366195,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002Fcb7e282912\u002Fhero-product-flow-05-finance.png",{},[521],{"top":522,"_uid":523,"left":524,"message":525,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"52","ff5f3180b5fc","40","Verifying shipment received...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":527,"is_external_url":198},{},[],{"_uid":530,"image":531,"pointer":535,"component":463,"image_mobile":540,"user_pointer":542},"81f2f2b62d89",{"id":532,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":533,"copyright":18,"fieldtype":19,"meta_data":534,"is_external_url":198},166009497374389,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F78fd54d669\u002Fhero-product-flow-06-finance.png",{},[536],{"top":537,"_uid":538,"left":461,"message":539,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"57","aba676516de9","Entering invoice details...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":541,"is_external_url":198},{},[],{"_uid":544,"image":545,"pointer":549,"component":463,"image_mobile":555,"user_pointer":557},"2740322b2645",{"id":546,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":547,"copyright":18,"fieldtype":19,"meta_data":548,"is_external_url":198},166009497362097,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002Fff0915e88d\u002Fhero-product-flow-07-finance.png",{},[550],{"top":551,"_uid":552,"left":553,"message":554,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"85","f95b5ca31017","72","Three-way match verified. Posting...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":556,"is_external_url":198},{},[],{"_uid":559,"image":560,"pointer":564,"component":463,"image_mobile":570,"user_pointer":572},"17bc13285f3e",{"id":561,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":562,"copyright":18,"fieldtype":19,"meta_data":563,"is_external_url":198},166009497407161,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F271124e761\u002Fhero-product-flow-08-finance.png",{},[565],{"top":566,"_uid":567,"left":568,"message":569,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"17","9f0472acdb13","16","Updating AP ledger...",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":571,"is_external_url":198},{},[],{"_uid":574,"image":575,"pointer":579,"component":463,"image_mobile":585,"user_pointer":587},"c0bcc286772b",{"id":576,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":577,"copyright":18,"fieldtype":19,"meta_data":578,"is_external_url":198},166009497386679,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1740x1024\u002F2a576026ed\u002Fhero-product-flow-09-finance.png",{},[580],{"top":581,"_uid":582,"left":583,"message":584,"component":459,"top_mobile":18,"left_mobile":18,"mobile_message":462},"77","b9c95638b994","29","Invoice reconciled!",{"id":465,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":466,"copyright":18,"fieldtype":19,"meta_data":586,"is_external_url":198},{},[],"How we build self-improving AI systems →","section-home-hero",{"type":28,"content":591},[592],{"type":31,"attrs":593,"content":594},{"textAlign":17},[595],{"text":437,"type":36},[597],{"_uid":598,"image":599,"component":603,"shader_config":604},"4659d151ffe2",{"id":600,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":601,"copyright":18,"fieldtype":19,"meta_data":602,"is_external_url":198},166268384716487,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F672x380\u002F0c1cc0ba09\u002Fhomepage-hero-shader-image.png",{},"shard-image",{"id":605,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":606,"copyright":18,"fieldtype":19,"meta_data":607,"is_external_url":198},166264362284558,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F52fbaf89fb\u002Fhomepage-hero-shader-config.json",{},{"id":125,"url":18,"linktype":24,"fieldtype":25,"cached_url":126,"story":609},{"name":128,"id":129,"uuid":125,"slug":130,"url":126,"full_slug":126,"_stopResolving":109},{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":611,"copyright":18,"fieldtype":19,"meta_data":612,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1344x498\u002F89f609aee5\u002Fbackground.png",{},{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":614,"copyright":18,"fieldtype":19,"meta_data":615,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F7113eff605\u002Fhomepage-hero-shader-config.json",{},{"_uid":617,"title":618,"brands":632,"eyebrow":642,"component":643},"d8fb20468180",{"type":28,"content":619},[620,627],{"type":31,"attrs":621,"content":622},{"textAlign":17},[623],{"text":624,"type":36,"marks":625},"testing",[626],{"type":183},{"type":31,"attrs":628,"content":629},{"textAlign":17},[630],{"text":631,"type":36},"test",[633],{"_uid":634,"logo":635,"image":637,"stats":639,"theme":640,"title":18,"callout":18,"component":641},"0a66ab94-b423-4d2e-ac31-e7814d61b67a",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":636},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":638},{},[],"light","showcase","Customers","section-home-showcase",{"_uid":645,"height":646,"to_color":647,"component":648,"from_color":649},"eaca9447d137","25vh","1","bg-transition","0",{"_uid":651,"cards":652,"title":817,"buttons":818,"eyebrow":823,"component":824,"hide_cards":109,"description":825,"shard_image":832},"2771814b9cb5",[653,684,716,757,787],{"_uid":654,"icon":655,"size":418,"eyebrow":659,"component":660,"description":661,"background_media":672,"background_media_mobile":683},"d4b556d2c8a4",{"id":656,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":657,"copyright":18,"fieldtype":19,"meta_data":658,"is_external_url":198},166269354596072,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F5c3319cce1\u002Ficon-bars-white.svg",{},"Ground truth","retention-card",{"type":28,"content":662},[663],{"type":31,"attrs":664,"content":665},{"textAlign":17},[666,670],{"text":667,"type":36,"marks":668},"Begin with correct.",[669],{"type":183},{"text":671,"type":36}," Pointer learns from your procedures, recordings, and the experts who run the process today, then encodes it all into living operating procedures. Ground truth exists before the first run starts.",[673],{"_uid":674,"image":675,"video":677,"lottie":679,"component":203},"7e0f64276976",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":676,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":678},{},{"id":680,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":681,"copyright":18,"fieldtype":19,"meta_data":682,"is_external_url":198},166270337324812,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F7df218cd35\u002Fproductmappinggatherinsights_opt.lottie",{},[],{"_uid":685,"icon":686,"size":418,"eyebrow":690,"component":660,"description":691,"background_media":704,"background_media_mobile":715},"24420352c430",{"id":687,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":688,"copyright":18,"fieldtype":19,"meta_data":689,"is_external_url":198},166269354591975,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F44bb0275f8\u002Ficon-up-white.svg",{},"ANY SYSTEM, ANY ERA",{"type":28,"content":692},[693,702],{"type":31,"attrs":694,"content":695},{"textAlign":17},[696,700],{"text":697,"type":36,"marks":698},"Runs any software, with or without an API.",[699],{"type":183},{"text":701,"type":36}," Most enterprise systems were never built to talk to each other. Pointer uses the screen the same way a human does, so it can reach anything a human can.",{"type":31,"attrs":703},{"textAlign":17},[705],{"_uid":706,"image":707,"video":709,"lottie":711,"component":203},"921b32d5f756",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":708,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":710},{},{"id":712,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":713,"copyright":18,"fieldtype":19,"meta_data":714,"is_external_url":198},166270135785213,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F18b339f71f\u002Fexecuteeverywhere_opt.lottie",{},[],{"_uid":717,"icon":718,"size":722,"eyebrow":723,"component":660,"description":724,"background_media":735,"background_media_mobile":746},"aef59c08-0190-4f58-ba25-39c8476bf486",{"id":719,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":720,"copyright":18,"fieldtype":19,"meta_data":721,"is_external_url":198},166269354563298,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002F8161a11b88\u002Ficon-radial-white.svg",{},"fullwidth","NOTHING ADVANCES UNVERIFIED",{"type":28,"content":725},[726],{"type":31,"attrs":727,"content":728},{"textAlign":17},[729,733],{"text":730,"type":36,"marks":731},"Prove every action.",[732],{"type":183},{"text":734,"type":36}," Each step is held against your standard of correct before the next begins. What can't be proven stops. Your team sees exceptions, not reviews.",[736],{"_uid":737,"image":738,"video":740,"lottie":742,"component":203},"4c1449f3-1552-4a59-8888-45109da2d0e9",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":739},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":741},{},{"id":743,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":744,"copyright":18,"fieldtype":19,"meta_data":745,"is_external_url":198},190366415532151,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F7c78bd4496\u002Fcontrolwithoversight_opt-extended-fixed-2-fps.lottie",{},[747],{"_uid":748,"image":749,"video":751,"lottie":753,"component":203},"dd3b37f3-9f67-480c-8c03-e354f163f1c0",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":750},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":752},{},{"id":754,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":755,"copyright":18,"fieldtype":19,"meta_data":756,"is_external_url":198},188936855534291,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F4aa79efa83\u002Fcontrolwithoversight_opt-restored-small.lottie",{},{"_uid":758,"icon":759,"size":418,"eyebrow":763,"component":660,"description":764,"background_media":775,"background_media_mobile":786},"bb5139f5ba79",{"id":760,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":761,"copyright":18,"fieldtype":19,"meta_data":762,"is_external_url":198},166269354542817,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002Ff934e97233\u002Ficon-circle-white.svg",{},"DETERMINISTIC BY DEFAULT",{"type":28,"content":765},[766],{"type":31,"attrs":767,"content":768},{"textAlign":17},[769,773],{"text":770,"type":36,"marks":771},"Run the known path, exactly. ",[772],{"type":183},{"text":774,"type":36},"When the process behaves as expected, Pointer executes the same steps the same way, every time. When it doesn't, intelligence steps in and finds the way back.",[776],{"_uid":777,"image":778,"video":780,"lottie":782,"component":203},"efc6640f18fd",{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":779},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":781},{},{"id":783,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":784,"copyright":18,"fieldtype":19,"meta_data":785,"is_external_url":198},166270135789310,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002Ffb0344cd19\u002Fadaptautomatically_opt.lottie",{},[],{"_uid":788,"icon":789,"size":418,"eyebrow":793,"component":660,"description":794,"background_media":805,"background_media_mobile":816},"f36149996d09",{"id":790,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":791,"copyright":18,"fieldtype":19,"meta_data":792,"is_external_url":198},166269354587877,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F28x28\u002Fd83add49f9\u002Ficon-explode-white.svg",{},"END TO END",{"type":28,"content":795},[796],{"type":31,"attrs":797,"content":798},{"textAlign":17},[799,803],{"text":800,"type":36,"marks":801},"Carry the whole process.",[802],{"type":183},{"text":804,"type":36}," The processes that hold a business together run for days, across departments and systems that were never built to talk to each other. Pointer runs them in the background, start to finish.",[806],{"_uid":807,"image":808,"video":810,"lottie":812,"component":203},"954a9f9e5c5f",{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":809},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":811},{},{"id":813,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":814,"copyright":18,"fieldtype":19,"meta_data":815,"is_external_url":198},188629150114439,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F325aed0b5a\u002Fscaleacrossusecases_opt-updated.lottie",{},[],"Built for the work agents can't hold.",[819],{"url":820,"_uid":822,"label":159,"component":256},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":821},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"40bc5fe75ba2","A new class of system","section-home-retention",{"type":28,"content":826},[827],{"type":31,"attrs":828,"content":829},{"textAlign":17},[830],{"text":831,"type":36},"Every company runs on a layer of work too critical to get wrong and too tangled for any tool to touch, so it stays with the people who hold it in their heads. Pointer is built to run that work on its own.",[833],{"_uid":834,"image":835,"component":603,"shader_config":838},"a108172034fb",{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":836,"copyright":18,"fieldtype":19,"meta_data":837,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F672x261\u002F7af3de76d9\u002Fhomepage-capabilities-shader-image-darkmode-min-min.png",{},{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":839,"copyright":18,"fieldtype":19,"meta_data":840,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F8df648a5ea\u002Fhomepage-capabilities-shader-config.json",{},{"_uid":842,"list":843,"media":898,"theme":907,"title":908,"button":926,"eyebrow":932,"component":933,"direction":418,"__home_switchback":109},"dccf66484465",[844],{"_uid":845,"items":846,"component":897},"d88c501b2cd7",[847,864,881],{"_uid":848,"title":18,"content":849,"component":860,"description":861},"ef9e65f947af",{"type":28,"content":850},[851],{"type":31,"attrs":852,"content":853},{"textAlign":17},[854,858],{"text":855,"type":36,"marks":856},"No system out of reach.",[857],{"type":183},{"text":859,"type":36}," Anything a person can do on a screen, Pointer can do – even on apps without APIs.","list-item",{"type":28,"content":862},[863],{"type":31},{"_uid":865,"title":18,"content":866,"component":860,"description":878},"6121d056b41c",{"type":28,"content":867},[868],{"type":31,"content":869},[870,874,876],{"text":871,"type":36,"marks":872},"More accurate than a human.",[873],{"type":183},{"text":875,"type":36}," ",{"text":877,"type":36},"Pointer reads the screen and acts on it, pulling the right figure and entering it in the right place across messy, unlabeled applications.",{"type":28,"content":879},[880],{"type":31},{"_uid":882,"title":18,"content":883,"component":860,"description":894},"5b47878bdc3e",{"type":28,"content":884},[885],{"type":31,"attrs":886,"content":887},{"textAlign":17},[888,892],{"text":889,"type":36,"marks":890},"Runs on its own in the background.",[891],{"type":183},{"text":893,"type":36}," Work executes in the cloud without a person driving it, so your team gets back the hours that went into steering and can spend them on the judgment calls that need them.",{"type":28,"content":895},[896],{"type":31},"bullet-list",[899],{"_uid":900,"image":901,"video":903,"lottie":905,"component":203},"5781c006de37",{"filename":17,"fieldtype":19,"meta_data":902},{},{"filename":18,"fieldtype":19,"meta_data":904},{},{"filename":713,"fieldtype":19,"meta_data":906},{},"dark",{"type":28,"content":909},[910,919],{"type":911,"attrs":912,"content":914},"heading",{"level":913,"textAlign":17},2,[915],{"text":916,"type":36,"marks":917},"Run the work, ",[918],{"type":183},{"type":911,"attrs":920,"content":921},{"level":913,"textAlign":17},[922],{"text":923,"type":36,"marks":924},"anywhere it lives.",[925],{"type":183},[927],{"url":928,"_uid":930,"label":931,"component":256},{"id":103,"url":18,"linktype":24,"fieldtype":25,"cached_url":104,"story":929},{"name":106,"id":107,"uuid":103,"slug":108,"url":104,"full_slug":104,"_stopResolving":109},"ed2158858ca1","More on computer use →","[GOES WHERE THE WORK IS]","section-switch-back",{"_uid":935,"list":936,"media":989,"theme":907,"title":998,"button":1007,"eyebrow":1008,"component":933,"direction":1009,"__home_switchback":109},"78af75665724",[937],{"_uid":938,"items":939,"component":897},"86e1f2ef282a",[940,956,972],{"_uid":941,"title":18,"content":942,"component":860,"description":953},"db6f2c88ea42",{"type":28,"content":943},[944],{"type":31,"content":945},[946,950,951],{"text":947,"type":36,"marks":948},"Each step is verified in the moment",[949],{"type":183},{"text":875,"type":36},{"text":952,"type":36},"the work happens, not at the end of the run, so a misstep stops instead of compounding across the next thousand.",{"type":28,"content":954},[955],{"type":31},{"_uid":957,"title":18,"content":958,"component":860,"description":969},"ab9ccb70b9c9",{"type":28,"content":959},[960],{"type":31,"content":961},[962,966,967],{"text":963,"type":36,"marks":964},"Measured against your standard of correct.",[965],{"type":183},{"text":875,"type":36},{"text":968,"type":36},"The work is checked against your company's specific definition of right rather than a general model's sense of what looks plausible.",{"type":28,"content":970},[971],{"type":31},{"_uid":973,"title":18,"content":974,"component":860,"description":986},"6c69a605e298",{"type":28,"content":975},[976],{"type":31,"attrs":977,"content":978},{"textAlign":17},[979,983,984],{"text":980,"type":36,"marks":981},"A checking system, not a second opinion.",[982],{"type":183},{"text":875,"type":36},{"text":985,"type":36},"Deterministic and stochastic checks run independently from execution, breaking a standard like \"is this expense reasonable\" into narrow questions each can verify.",{"type":28,"content":987},[988],{"type":31},[990],{"_uid":991,"image":992,"video":994,"lottie":996,"component":203},"a1864ddfc99c",{"filename":17,"fieldtype":19,"meta_data":993},{},{"filename":18,"fieldtype":19,"meta_data":995},{},{"filename":744,"fieldtype":19,"meta_data":997},{},{"type":28,"content":999},[1000],{"type":911,"attrs":1001,"content":1002},{"level":913,"textAlign":17},[1003],{"text":1004,"type":36,"marks":1005},"Nothing advances until it's proven.",[1006],{"type":183},[],"[PROVES EVERY STEP]","reverse",{"_uid":1011,"list":1012,"media":1066,"theme":907,"title":1075,"button":1090,"eyebrow":1096,"component":933,"direction":418,"__home_switchback":109},"dec730f83552",[1013],{"_uid":1014,"items":1015,"component":897},"61dbe67955ad",[1016,1032,1048],{"_uid":1017,"title":18,"content":1018,"component":860,"description":1029},"434047416c3e",{"type":28,"content":1019},[1020],{"type":31,"content":1021},[1022,1026,1027],{"text":1023,"type":36,"marks":1024},"Proven reasoning hardens into code.",[1025],{"type":183},{"text":875,"type":36},{"text":1028,"type":36},"Each run converts more of the work from live reasoning to fixed code, so drift falls, cost drops, and certainty climbs.",{"type":28,"content":1030},[1031],{"type":31},{"_uid":1033,"title":18,"content":1034,"component":860,"description":1045},"92664218a433",{"type":28,"content":1035},[1036],{"type":31,"attrs":1037,"content":1038},{"textAlign":17},[1039,1043],{"text":1040,"type":36,"marks":1041},"Built for long-horizon finance and operations processes ",[1042],{"type":183},{"text":1044,"type":36},"that run for hours or days across systems and departments, holding their path without a person steering each turn.",{"type":28,"content":1046},[1047],{"type":31},{"_uid":1049,"title":18,"content":1050,"component":860,"description":1063},"050d4f1be48f",{"type":28,"content":1051},[1052,1061],{"type":31,"attrs":1053,"content":1054},{"textAlign":17},[1055,1059],{"text":1056,"type":36,"marks":1057},"Only the novel reaches a person.",[1058],{"type":183},{"text":1060,"type":36}," The process runs unattended and routes out only what's genuinely new, and each exception your team resolves becomes a proven path it runs on its own next time.",{"type":31,"attrs":1062},{"textAlign":17},{"type":28,"content":1064},[1065],{"type":31},[1067],{"_uid":1068,"image":1069,"video":1071,"lottie":1073,"component":203},"d3e1f64689be",{"filename":17,"fieldtype":19,"meta_data":1070},{},{"filename":18,"fieldtype":19,"meta_data":1072},{},{"filename":814,"fieldtype":19,"meta_data":1074},{},{"type":28,"content":1076},[1077],{"type":911,"attrs":1078,"content":1079},{"level":913,"textAlign":17},[1080,1084,1086],{"text":1081,"type":36,"marks":1082},"More certain with ",[1083],{"type":183},{"type":1085},"hard_break",{"text":1087,"type":36,"marks":1088},"every run.",[1089],{"type":183},[1091],{"url":1092,"_uid":1094,"label":1095,"component":256},{"id":125,"url":18,"linktype":24,"fieldtype":25,"cached_url":126,"story":1093},{"name":128,"id":129,"uuid":125,"slug":130,"url":126,"full_slug":126,"_stopResolving":109},"52afafb9-80c7-4dc9-ae19-f790b23d2dc7","More on self-improvement →","[HOLDS OVER TIME]",{"_uid":1098,"height":646,"to_color":649,"component":648,"from_color":647},"06106925f661",{"_uid":1100,"text":1101,"anchor":18,"images":1135,"buttons":1144,"entries":1154,"component":1155,"shard_config":1156},"6a203d5c2d6a",{"type":28,"content":1102},[1103,1108,1120,1125],{"type":31,"attrs":1104,"content":1105},{"textAlign":17},[1106],{"text":1107,"type":36},"Most AI is made safe by ",{"type":31,"attrs":1109,"content":1110},{"textAlign":17},[1111,1118],{"text":1112,"type":36,"marks":1113},"limiting",[1114],{"type":1115,"attrs":1116},"highlight",{"color":1117},"#7E8BFF",{"text":1119,"type":36}," what it can do. ",{"type":31,"attrs":1121,"content":1122},{"textAlign":17},[1123],{"text":1124,"type":36},"Pointer is made safe by ",{"type":31,"attrs":1126,"content":1127},{"textAlign":17},[1128,1133],{"text":1129,"type":36,"marks":1130},"verifying",[1131],{"type":1115,"attrs":1132},{"color":1117},{"text":1134,"type":36}," everything it does.",[1136,1140],{"id":1137,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1138,"copyright":18,"fieldtype":19,"meta_data":1139},166268384708294,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F774x722\u002Fe2c027983b\u002Fhomepage-blurb-1-shader-image.png",{},{"id":1141,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1142,"copyright":18,"fieldtype":19,"meta_data":1143},166268384720584,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F774x722\u002F626d00c65f\u002Fhomepage-blurb-2-shader-image.png",{},[1145,1149],{"url":1146,"_uid":1148,"type":242,"label":159,"component":243},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":1147},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"63a01b915263",{"url":1150,"_uid":1152,"type":349,"label":1153,"component":243},{"id":18,"url":1151,"linktype":172,"fieldtype":25,"cached_url":1151},"https:\u002F\u002Ftrust.pointer.ai","78fdd0b2e921","View trust center",[],"section-home-blurb",{"id":1157,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1158,"copyright":18,"fieldtype":19,"meta_data":1159,"is_external_url":198},166264362243593,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F9e269adad7\u002Fhomepage-blurb-1-shader-config.json",{},{"_uid":1161,"entries":1162,"component":1234},"bdcfe009990d",[1163,1171,1178,1185,1197,1204,1220,1227],{"_uid":1164,"logo":1165,"eyebrow":1169,"component":1170},"7af43dfa161d",{"id":1166,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1167,"copyright":18,"fieldtype":19,"meta_data":1168,"is_external_url":198},165982603504846,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F189x189\u002F629641e535\u002Fhipaa.png",{},"HIPAA","card-data-logo",{"_uid":1172,"logo":1173,"eyebrow":1177,"component":1170},"7d47d68db6cd",{"id":1174,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1175,"copyright":18,"fieldtype":19,"meta_data":1176,"is_external_url":198},165985075514579,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F378x378\u002Fa95d3deebf\u002Fsoc2-badge.png",{},"SOC2",{"_uid":1179,"logo":1180,"eyebrow":1184,"component":1170},"eb56d490391b",{"id":1181,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1182,"copyright":18,"fieldtype":19,"meta_data":1183,"is_external_url":198},165985182411990,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F369x369\u002F9280ca0949\u002Fsecurity-badge-aiuc.png",{},"AIUC",{"_uid":1186,"color":1187,"title":1188,"content":1189,"component":1196},"26b0a93630d8","orange","Transparent by design.",{"type":28,"content":1190},[1191],{"type":31,"attrs":1192,"content":1193},{"textAlign":17},[1194],{"text":1195,"type":36},"Every step is recorded against the standard it was checked on, so the work holds up when an auditor, a regulator, or your own controls team asks how it was done.","card-data-content",{"_uid":1198,"logo":1199,"color":1201,"quote":1202,"author":18,"component":1203},"8f300282038e",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1200,"is_external_url":198},{},"purple","Designed for sensitive production systems, with SSO, RBAC, encryption, audit logs, VPC support, and configurable data controls.","card-data-quote",{"_uid":1205,"color":1201,"label":1206,"value":1207,"component":1208,"background_media":1209},"83fc3ee33bff","uptime sla","99.9%","card-data-stat",[1210],{"_uid":1211,"image":1212,"video":1214,"lottie":1218,"component":203},"5afe3c73a65f",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1213,"is_external_url":198},{},{"id":1215,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1216,"copyright":18,"fieldtype":19,"meta_data":1217,"is_external_url":198},166268470150859,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F0ff8e0f7bb\u002Fpointer-person-background-vid-loop-compressed.mp4",{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1219},{},{"_uid":1221,"logo":1222,"eyebrow":1226,"component":1170},"67063aa870d6",{"id":1223,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1224,"copyright":18,"fieldtype":19,"meta_data":1225,"is_external_url":198},165982603508943,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F189x189\u002Fb7816a2b65\u002Fgdpr.png",{},"GDPR",{"_uid":1228,"logo":1229,"eyebrow":1233,"component":1170},"298312c3adf8",{"id":1230,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1231,"copyright":18,"fieldtype":19,"meta_data":1232,"is_external_url":198},165984984300754,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F378x378\u002Fbeede5e1b0\u002Fiso-badge.png",{},"ISO 27001","section-home-wall-of-credibility",{"_uid":1236,"title":1237,"eyebrow":1249,"component":1250,"description":1251,"shard_image":1262,"background_image":1273},"a9533ce3b772",{"type":28,"content":1238},[1239,1244],{"type":31,"attrs":1240,"content":1241},{"textAlign":17},[1242],{"text":1243,"type":36},"Turn your operation into a ",{"type":31,"attrs":1245,"content":1246},{"textAlign":17},[1247],{"text":1248,"type":36},"system that runs itself.","How it works","section-home-header",{"type":28,"content":1252},[1253],{"type":31,"attrs":1254,"content":1255},{"textAlign":17},[1256],{"text":1257,"type":36,"marks":1258},"Our platform learns how your best people do the work, encodes it, proves it against your standard, and runs it on its own. We stand up the first processes with you, then your team makes it their own.",[1259],{"type":1260,"attrs":1261},"textStyle",{"color":18},[1263],{"_uid":1264,"image":1265,"component":603,"shader_config":1269},"3f1632c28a6f",{"id":1266,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1267,"copyright":18,"fieldtype":19,"meta_data":1268,"is_external_url":198},166268384704197,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F384x292\u002Fcd81934f52\u002Fhomepage-engine-shader-image.png",{},{"id":1270,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1271,"copyright":18,"fieldtype":19,"meta_data":1272,"is_external_url":198},166264362284557,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F360765d7b3\u002Fhomepage-engine-shader-config.json",{},{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1274,"copyright":18,"fieldtype":19,"meta_data":1275,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1534x1168\u002F2e8d850075\u002Fwebgl-shards.png",{},{"_uid":1277,"entries":1278,"component":1372},"e19293ae2113",[1279,1303,1326,1349],{"_uid":1280,"color":1201,"media":1281,"title":1292,"eyebrow":1293,"component":1294,"description":1295,"animation_duration":1302},"062de694ba39",[1282],{"_uid":1283,"image":1284,"video":1286,"lottie":1288,"component":203},"5ca5fad30157",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1285,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1287},{},{"id":1289,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1290,"copyright":18,"fieldtype":19,"meta_data":1291,"is_external_url":198},188292744472665,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002Fb17b5220a4\u002Fknowledgeintegrations_opt-2.lottie",{},"Learn how the work is done","Learn","product-card",{"type":28,"content":1296},[1297],{"type":31,"attrs":1298,"content":1299},{"textAlign":17},[1300],{"text":1301,"type":36},"Our deployment team sits with the people who run the process today and pull out how it actually works from SOPs, videos, historical data, and the hundred small calls your team makes every day. What comes out the other side is an explicit definition of correct for the process, the thing that only lived in your best people's heads until now.","22",{"_uid":1304,"color":1187,"media":1305,"title":1316,"eyebrow":1317,"component":1294,"description":1318,"animation_duration":1325},"55f76cf5b9d6",[1306],{"_uid":1307,"image":1308,"video":1310,"lottie":1314,"component":203},"83142d4433a1",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1309,"is_external_url":198},{},{"id":1311,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1312,"copyright":18,"fieldtype":19,"meta_data":1313,"is_external_url":198},188322263386543,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F1a3dbac861\u002Fproductmapping_opt-updated-2.mp4",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1315,"is_external_url":198},{},"Encode and evaluate","Encode",{"type":28,"content":1319},[1320],{"type":31,"attrs":1321,"content":1322},{"textAlign":17},[1323],{"text":1324,"type":36},"That definition becomes a procedure: your process in plain language, run by a mix of code and AI. Before it touches production, we put it through hundreds of edge and adversarial cases in simulation, the way code clears its tests before it ships. Now the process exists as something you own and can trust, not a practice that walks out the door when someone leaves.","11.5",{"_uid":1327,"color":1187,"media":1328,"title":1339,"eyebrow":1340,"component":1294,"description":1341,"animation_duration":1348},"73141bdecede",[1329],{"_uid":1330,"image":1331,"video":1333,"lottie":1335,"component":203},"9cbb66914ce8",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1332,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1334},{},{"id":1336,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1337,"copyright":18,"fieldtype":19,"meta_data":1338,"is_external_url":198},190363632156751,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002Fabbf457fb5\u002Fcontext-aware_opt-updated-2.lottie",{},"Run in production","Execute",{"type":28,"content":1342},[1343],{"type":31,"attrs":1344,"content":1345},{"textAlign":17},[1346],{"text":1347,"type":36},"The procedure runs in the background, across the systems that don't talk to each other, for as long as the work takes, with every step verified as it happens. The parts it has mastered run as fixed code, and only what's unfamiliar or failing reaches your team. What you have now is the work getting done on its own, with a record of every step behind it.","12.9",{"_uid":1350,"color":1201,"media":1351,"title":1362,"eyebrow":1363,"component":1294,"description":1364,"animation_duration":1371},"115125df92ef",[1352],{"_uid":1353,"image":1354,"video":1356,"lottie":1358,"component":203},"511e6f7af05e",{"alt":17,"name":18,"focus":17,"title":17,"filename":17,"copyright":17,"fieldtype":19,"meta_data":1355,"is_external_url":198},{},{"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1357},{},{"id":1359,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1360,"copyright":18,"fieldtype":19,"meta_data":1361,"is_external_url":198},166274451457944,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002F995eed4f48\u002Fbehavioral-feedback_opt.lottie",{},"Continuous learning","Sharpen",{"type":28,"content":1365},[1366],{"type":31,"attrs":1367,"content":1368},{"textAlign":17},[1369],{"text":1370,"type":36},"The system never stops learning. Every run, every exception, every correction from your experts makes it more attuned to how the process really works, and more of the work completes without any human intervention.","9","section-home-products",{"_uid":1374,"height":1375,"to_color":647,"component":648,"from_color":649},"29b0ee87094e","75vh",{"_uid":1377,"image":1378,"steps":1381,"title":1382,"eyebrow":1393,"component":1394,"description":1395},"f3a959d0bcbe",{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1379,"copyright":18,"fieldtype":19,"meta_data":1380,"is_external_url":198},"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F888x888\u002F88d920bac6\u002Fhow-to-start-clock.png",{},[],{"type":28,"content":1383},[1384],{"type":31,"attrs":1385,"content":1386},{"textAlign":17},[1387],{"text":1388,"type":36,"marks":1389},"Best-in-class performance.",[1390,1392],{"type":1260,"attrs":1391},{"color":18},{"type":183},"More on our research","section-home-get-started",{"type":28,"content":1396},[1397],{"type":31,"attrs":1398,"content":1399},{"textAlign":17},[1400],{"text":1401,"type":36},"Pointer holds the top score on the default benchmark that measures how capable a model is at doing computer tasks.",{"_uid":1403,"height":1375,"to_color":649,"component":648,"from_color":647},"af4bec971b11",{"_uid":1405,"image":1406,"title":1410,"button":1411,"eyebrow":400,"component":1416,"description":1417,"shard_image":1424,"background_image":1435},"a4a6f52d98dd",{"id":1407,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1408,"copyright":18,"fieldtype":19,"meta_data":1409,"is_external_url":198},166010666311454,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1956x604\u002Fc8f4a2ff7c\u002Ffinal-cta-product-finance-erp.png",{},"Partner with Pointer. ",[1412],{"url":1413,"_uid":1415,"label":159,"component":256},{"id":162,"url":18,"linktype":24,"fieldtype":25,"cached_url":163,"story":1414},{"name":165,"id":166,"uuid":162,"slug":163,"url":163,"full_slug":163,"_stopResolving":109},"23bfc736d2e8","section-pre-footer",{"type":28,"attrs":1418,"content":1419},{"backgroundColor":17},[1420],{"type":31,"attrs":1421,"content":1422},{"textAlign":17},[1423],{"text":277,"type":36},[1425],{"_uid":1426,"image":1427,"component":603,"shader_config":1431},"eecccef379b1",{"id":1428,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1429,"copyright":18,"fieldtype":19,"meta_data":1430,"is_external_url":198},166268384724681,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F672x249\u002F3920517719\u002Fhomepage-prefooter-shader-image.png",{},{"id":1432,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1433,"copyright":18,"fieldtype":19,"meta_data":1434,"is_external_url":198},166264362255882,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002Fx\u002Fe1108086da\u002Fhomepage-prefooter-shader-config.json",{},{"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":611,"copyright":18,"fieldtype":19,"meta_data":1436,"is_external_url":198},{},"home-template",-50,[],"373ac607-2d2c-4e31-bd8b-5ead667c2e06","2026-04-14T23:21:38.092Z",[],{"name":128,"created_at":1444,"published_at":1445,"updated_at":1446,"id":129,"uuid":125,"content":1447,"slug":130,"full_slug":126,"sort_by_date":17,"position":2241,"tag_list":2242,"is_startpage":198,"parent_id":2243,"meta_data":17,"group_id":2244,"first_published_at":2245,"release_id":17,"lang":418,"path":17,"alternates":2246,"default_full_slug":17,"translated_slugs":17},"2026-08-11T00:53:11.892Z","2026-08-15T19:44:32.250Z","2026-08-15T19:44:32.271Z",{"seo":1448,"_uid":1458,"body":1459,"image":2195,"title":2200,"topic":2209,"author":2210,"excerpt":2231,"component":2238,"publishedOn":2239,"hide_on_index":198,"related_articles":2240},[1449],{"_uid":1450,"image":1451,"title":1456,"component":436,"description":1457},"734ce4c6-73da-4ea0-b158-99a256898dbc",{"id":1452,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1453,"copyright":18,"fieldtype":19,"meta_data":1454,"is_external_url":198},209439738805254,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F2440x1374\u002F54585514a6\u002Fhydra-og.png",{"size":1455},"2440x1374","Pointer - Teaching AI systems to learn from their own work","We built a system that improves itself in production without a person behind it, in workflows where a single silent change can become a permanent bypass. ","aa4972a9-52a4-4e00-9c07-7fd71fcff8f2",[1460,1516,1559,1814,1831,1944,1958,1968,2020,2035,2050,2116],{"_uid":1461,"anchor":1462,"content":1463,"eyebrow":18,"component":1515},"37370b18-02f3-439d-9da4-11543e7f37ef","Learning without drift",{"type":28,"attrs":1464,"content":1465},{"backgroundColor":17},[1466,1471,1482,1487,1492,1497,1502],{"type":911,"attrs":1467,"content":1469},{"level":1468,"textAlign":17},4,[1470],{"text":1462,"type":36},{"type":1472,"content":1473},"blockquote",[1474],{"type":31,"attrs":1475,"content":1476},{"textAlign":17},[1477],{"text":1478,"type":36,"marks":1479},"“The only real mistake is the one from which we learn nothing” - Henry Ford",[1480],{"type":1481},"italic",{"type":31,"attrs":1483,"content":1484},{"textAlign":17},[1485],{"text":1486,"type":36},"Learning from mistakes over time is one of the many things that we humans are so great at that we take for granted. While it may seem trivial, rather complex computations and decision trees take shape in our heads any time we attempt a task that we have failed at before. We subconsciously keep track of what we tried that didn’t work, what did eventually work, and rather importantly, why it worked.",{"type":31,"attrs":1488,"content":1489},{"textAlign":17},[1490],{"text":1491,"type":36},"Unlike people, most production agents start each run virtually from scratch. Over the course of a run, they can recover from a bad tool call, discover that a document needs to be parsed a certain way, or even find a clever trick… and then discard that experience when the run ends. This allows subsequent runs to make the same mistakes.",{"type":31,"attrs":1493,"content":1494},{"textAlign":17},[1495],{"text":1496,"type":36},"At production scale, the same avoidable failure can recur across hundreds of invoices, claims, or reviews. Agents can scale repetition much faster than they scale learning; and without an improvement loop, they scale their mistakes too.",{"type":31,"attrs":1498,"content":1499},{"textAlign":17},[1500],{"text":1501,"type":36},"Memory seems to be the obvious answer: let the agent remember everything and teach itself as it goes. Unchecked memory that changes future behavior, however, is no longer just memory. It becomes policy. A supplier might be exempt from one approval because of a temporary exception. If the agent remembers only “this supplier does not require approval,” a one-time recovery can silently become a permanent bypass. To avoid such issues, policy, in production, must have evidence, and be scoped, reviewable, and reversible.",{"type":31,"attrs":1503,"content":1504},{"textAlign":17},[1505,1507,1513],{"text":1506,"type":36},"We built our self-improvement system, called “Hydra” internally, around this constraint. It turns execution evidence into proposed improvements while maintaining these standards. Like its ",{"text":1508,"type":36,"marks":1509},"Lernaean namesake",[1510],{"type":256,"attrs":1511},{"href":1512,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FLernaean_Hydra",{"text":1514,"type":36},", the system comes back stronger every time.","chapter",{"_uid":1517,"anchor":1518,"content":1519,"eyebrow":18,"component":1515},"5936fd86-0cb5-4673-ae6d-1a6d847cf774","How we measured it",{"type":28,"attrs":1520,"content":1521},{"backgroundColor":17},[1522,1526,1531,1536,1541,1554],{"type":911,"attrs":1523,"content":1524},{"level":1468,"textAlign":17},[1525],{"text":1518,"type":36},{"type":31,"attrs":1527,"content":1528},{"textAlign":17},[1529],{"text":1530,"type":36},"To measure Hydra, we ran it against a fixed set of work that mirrors what Pointer runs in production. The unit of that work is a procedure: a recurring operational task with a right answer, akin to a standard operating procedure. Think a close reconciliation, invoice auditing, or clearing chargebacks.",{"type":31,"attrs":1532,"content":1533},{"textAlign":17},[1534],{"text":1535,"type":36},"For each one we assembled scenarios, each a fixed set of inputs with a known correct result, carrying the same mixture of emails, attachments, spreadsheets, business rules, judgment calls, and incomplete instructions that is standard in real operations.",{"type":31,"attrs":1537,"content":1538},{"textAlign":17},[1539],{"text":1540,"type":36},"We built these scenarios to be difficult on purpose. They skew toward the adversarial cases we expect in production rather than the clean ones. Corrupt documents, ambiguous decisions, threshold calculations, competing policy rules, human review requirements, inputs scattered across files and applications. Some require the runtime to recognize that it cannot safely proceed.",{"type":31,"attrs":1542,"content":1543},{"textAlign":17},[1544,1546,1552],{"text":1545,"type":36},"The procedures were realistic in another important way - their SOPs were useful, but not exhaustive or foolproof. That is how production work is usually specified. An SOP defines the work but it rarely anticipates every malformed attachment, conflicting field, missing integration, edge condition, or judgment call. They are like ",{"text":1547,"type":36,"marks":1548},"Jasper O’Farrell’s map of San Francisco",[1549],{"type":256,"attrs":1550},{"href":1551,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fwww.sfomuseum.org\u002Fsites\u002Fdefault\u002Ffiles\u002F01_95.jpg",{"text":1553,"type":36},", in that they lay out what needs to be done in broad strokes but don’t guarantee that the terrain will cooperate.",{"type":31,"attrs":1555,"content":1556},{"textAlign":17},[1557],{"text":1558,"type":36},"The result is that the opening numbers reflect cold performance against the worst of production, not typical performance. Out of the box means the runtime executes each procedure from its SOP as written, with no findings active. We assembled 68 scenarios across 10 procedures and ran three rounds, feeding each round's evidence to Hydra before the next. Scenarios within a procedure were distinct work items that shared failure types, not answers, so a finding extracted in one round could only improve the next by generalizing to work it had never seen.",{"_uid":1560,"anchor":1561,"content":1562,"eyebrow":18,"component":1515},"fbaff684-b0e8-4073-8fc7-13ff0535d6d9","More work, done correctly",{"type":28,"attrs":1563,"content":1564},{"backgroundColor":17},[1565,1569,1574,1579,1627,1632,1809],{"type":911,"attrs":1566,"content":1567},{"level":1468,"textAlign":17},[1568],{"text":1561,"type":36},{"type":31,"attrs":1570,"content":1571},{"textAlign":17},[1572],{"text":1573,"type":36},"At Pointer, the result that we care the most about is correctness. We work in spaces where being almost right is the same as being wrong.",{"type":31,"attrs":1575,"content":1576},{"textAlign":17},[1577],{"text":1578,"type":36},"We measured several dimensions separately because a single “success rate” hides too much:",{"type":1580,"content":1581},"bullet_list",[1582,1590,1613,1620],{"type":1583,"content":1584},"list_item",[1585],{"type":31,"attrs":1586,"content":1587},{"textAlign":17},[1588],{"text":1589,"type":36},"Operational completion rate: measures how often a run reached a completed state. The first requirement for winning a race is to get to the finish line.",{"type":1583,"content":1591},[1592,1597],{"type":31,"attrs":1593,"content":1594},{"textAlign":17},[1595],{"text":1596,"type":36},"Task-critical accuracy: combines the scored fields within each scenario and balances the result across procedures. It includes:",{"type":1580,"content":1598},[1599,1606],{"type":1583,"content":1600},[1601],{"type":31,"attrs":1602,"content":1603},{"textAlign":17},[1604],{"text":1605,"type":36},"Terminal-decision accuracy: whether the runtime made the correct final decision for each work item.",{"type":1583,"content":1607},[1608],{"type":31,"attrs":1609,"content":1610},{"textAlign":17},[1611],{"text":1612,"type":36},"Numerical accuracy: task-critical amounts, ratios, thresholds, and normalized values.",{"type":1583,"content":1614},[1615],{"type":31,"attrs":1616,"content":1617},{"textAlign":17},[1618],{"text":1619,"type":36},"Cost per completion",{"type":1583,"content":1621},[1622],{"type":31,"attrs":1623,"content":1624},{"textAlign":17},[1625],{"text":1626,"type":36},"Median run duration",{"type":31,"attrs":1628,"content":1629},{"textAlign":17},[1630],{"text":1631,"type":36},"Every quality metric climbed across the three rounds:",{"type":1633,"content":1634},"table",[1635,1672,1707,1741,1775],{"type":1636,"content":1637},"tableRow",[1638,1648,1656,1664],{"type":1639,"attrs":1640,"content":1642},"tableHeader",{"colspan":1641,"rowspan":1641,"colwidth":17},1,[1643],{"type":31,"attrs":1644,"content":1645},{"textAlign":17},[1646],{"text":1647,"type":36},"Metric",{"type":1639,"attrs":1649,"content":1650},{"colspan":1641,"rowspan":1641,"colwidth":17},[1651],{"type":31,"attrs":1652,"content":1653},{"textAlign":17},[1654],{"text":1655,"type":36},"Out of the box",{"type":1639,"attrs":1657,"content":1658},{"colspan":1641,"rowspan":1641,"colwidth":17},[1659],{"type":31,"attrs":1660,"content":1661},{"textAlign":17},[1662],{"text":1663,"type":36},"After Hydra",{"type":1639,"attrs":1665,"content":1666},{"colspan":1641,"rowspan":1641,"colwidth":17},[1667],{"type":31,"attrs":1668,"content":1669},{"textAlign":17},[1670],{"text":1671,"type":36},"Change",{"type":1636,"content":1673},[1674,1683,1691,1699],{"type":1675,"attrs":1676,"content":1677},"tableCell",{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1678],{"type":31,"attrs":1679,"content":1680},{"textAlign":17},[1681],{"text":1682,"type":36},"Task-critical accuracy",{"type":1675,"attrs":1684,"content":1685},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1686],{"type":31,"attrs":1687,"content":1688},{"textAlign":17},[1689],{"text":1690,"type":36},"54.4%",{"type":1675,"attrs":1692,"content":1693},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1694],{"type":31,"attrs":1695,"content":1696},{"textAlign":17},[1697],{"text":1698,"type":36},"76.8%",{"type":1675,"attrs":1700,"content":1701},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1702],{"type":31,"attrs":1703,"content":1704},{"textAlign":17},[1705],{"text":1706,"type":36},"+22.4 pp",{"type":1636,"content":1708},[1709,1717,1725,1733],{"type":1675,"attrs":1710,"content":1711},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1712],{"type":31,"attrs":1713,"content":1714},{"textAlign":17},[1715],{"text":1716,"type":36},"Terminal decision accuracy",{"type":1675,"attrs":1718,"content":1719},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1720],{"type":31,"attrs":1721,"content":1722},{"textAlign":17},[1723],{"text":1724,"type":36},"46.4%",{"type":1675,"attrs":1726,"content":1727},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1728],{"type":31,"attrs":1729,"content":1730},{"textAlign":17},[1731],{"text":1732,"type":36},"89.1%",{"type":1675,"attrs":1734,"content":1735},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1736],{"type":31,"attrs":1737,"content":1738},{"textAlign":17},[1739],{"text":1740,"type":36},"+42.7 pp",{"type":1636,"content":1742},[1743,1751,1759,1767],{"type":1675,"attrs":1744,"content":1745},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1746],{"type":31,"attrs":1747,"content":1748},{"textAlign":17},[1749],{"text":1750,"type":36},"Numerical accuracy",{"type":1675,"attrs":1752,"content":1753},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1754],{"type":31,"attrs":1755,"content":1756},{"textAlign":17},[1757],{"text":1758,"type":36},"28.6%",{"type":1675,"attrs":1760,"content":1761},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1762],{"type":31,"attrs":1763,"content":1764},{"textAlign":17},[1765],{"text":1766,"type":36},"92.9%",{"type":1675,"attrs":1768,"content":1769},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1770],{"type":31,"attrs":1771,"content":1772},{"textAlign":17},[1773],{"text":1774,"type":36},"+64.3 pp",{"type":1636,"content":1776},[1777,1785,1793,1801],{"type":1675,"attrs":1778,"content":1779},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1780],{"type":31,"attrs":1781,"content":1782},{"textAlign":17},[1783],{"text":1784,"type":36},"Operational completion rate",{"type":1675,"attrs":1786,"content":1787},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1788],{"type":31,"attrs":1789,"content":1790},{"textAlign":17},[1791],{"text":1792,"type":36},"79.4%",{"type":1675,"attrs":1794,"content":1795},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1796],{"type":31,"attrs":1797,"content":1798},{"textAlign":17},[1799],{"text":1800,"type":36},"100%",{"type":1675,"attrs":1802,"content":1803},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1804],{"type":31,"attrs":1805,"content":1806},{"textAlign":17},[1807],{"text":1808,"type":36},"+20.6 pp",{"type":31,"attrs":1810,"content":1811},{"textAlign":17},[1812],{"text":1813,"type":36},"The runtime finished more often, made better final decisions, and produced substantially more accurate numbers. Accuracy is what decides whether the work can run without a person behind it. Every decision the runtime gets right is one nobody has to review, and every one it gets wrong is rework, so these numbers are the difference between a system that removes work and a system that creates more.",{"_uid":1815,"image":1816,"layout":1828,"caption":1829,"component":1830},"a452c0d9-90ed-4127-bff8-2f1a162ae77a",[1817],{"_uid":1818,"image":1819,"video":1824,"lottie":1826,"component":203},"509f35e3-fc07-4296-b1b5-8cf348b496d9",{"id":1820,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1821,"copyright":18,"fieldtype":19,"meta_data":1822,"is_external_url":198},207774751179880,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2176\u002Ffde6906a4d\u002Fhydra-accuracy.png",{"size":1823},"3760x2176",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1825},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1827},{},"block","Accuracy raises round over round: decisions, numbers, and completions all climb as the system learns from its own runs. Higher is better.","article-image-block",{"_uid":1832,"anchor":18,"content":1833,"eyebrow":18,"component":1515},"8178aba1-ec06-473e-8bc7-4e5f10527524",{"type":28,"attrs":1834,"content":1835},{"backgroundColor":17},[1836,1841,1939],{"type":31,"attrs":1837,"content":1838},{"textAlign":17},[1839],{"text":1840,"type":36},"A system that gets faster at being wrong is not improving, so the movement that matters happened on both sides of the ledger. While quality climbed, the cost of each completed run fell, and the runs got faster:",{"type":1633,"content":1842},[1843,1873,1906],{"type":1636,"content":1844},[1845,1852,1859,1866],{"type":1639,"attrs":1846,"content":1847},{"colspan":1641,"rowspan":1641,"colwidth":17},[1848],{"type":31,"attrs":1849,"content":1850},{"textAlign":17},[1851],{"text":1647,"type":36},{"type":1639,"attrs":1853,"content":1854},{"colspan":1641,"rowspan":1641,"colwidth":17},[1855],{"type":31,"attrs":1856,"content":1857},{"textAlign":17},[1858],{"text":1655,"type":36},{"type":1639,"attrs":1860,"content":1861},{"colspan":1641,"rowspan":1641,"colwidth":17},[1862],{"type":31,"attrs":1863,"content":1864},{"textAlign":17},[1865],{"text":1663,"type":36},{"type":1639,"attrs":1867,"content":1868},{"colspan":1641,"rowspan":1641,"colwidth":17},[1869],{"type":31,"attrs":1870,"content":1871},{"textAlign":17},[1872],{"text":1671,"type":36},{"type":1636,"content":1874},[1875,1882,1890,1898],{"type":1675,"attrs":1876,"content":1877},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1878],{"type":31,"attrs":1879,"content":1880},{"textAlign":17},[1881],{"text":1619,"type":36},{"type":1675,"attrs":1883,"content":1884},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1885],{"type":31,"attrs":1886,"content":1887},{"textAlign":17},[1888],{"text":1889,"type":36},"$4.28",{"type":1675,"attrs":1891,"content":1892},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1893],{"type":31,"attrs":1894,"content":1895},{"textAlign":17},[1896],{"text":1897,"type":36},"$3.47",{"type":1675,"attrs":1899,"content":1900},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1901],{"type":31,"attrs":1902,"content":1903},{"textAlign":17},[1904],{"text":1905,"type":36},"−18.9%",{"type":1636,"content":1907},[1908,1915,1923,1931],{"type":1675,"attrs":1909,"content":1910},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1911],{"type":31,"attrs":1912,"content":1913},{"textAlign":17},[1914],{"text":1626,"type":36},{"type":1675,"attrs":1916,"content":1917},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1918],{"type":31,"attrs":1919,"content":1920},{"textAlign":17},[1921],{"text":1922,"type":36},"501 sec",{"type":1675,"attrs":1924,"content":1925},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1926],{"type":31,"attrs":1927,"content":1928},{"textAlign":17},[1929],{"text":1930,"type":36},"339 sec",{"type":1675,"attrs":1932,"content":1933},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[1934],{"type":31,"attrs":1935,"content":1936},{"textAlign":17},[1937],{"text":1938,"type":36},"−32.3%",{"type":31,"attrs":1940,"content":1941},{"textAlign":17},[1942],{"text":1943,"type":36},"Cost is what decides whether any of this runs at production volume. Accuracy that arrives by spending more on compute and retries stops making sense at thousands of runs a month. Here the system got more accurate and cheaper at the same time, because much of what it learned was how to stop wasting effort.",{"_uid":1945,"image":1946,"layout":1828,"caption":1957,"component":1830},"be1b2619-1347-49ff-8d62-dd43dbe43796",[1947],{"_uid":1948,"image":1949,"video":1953,"lottie":1955,"component":203},"c5766dff-5650-4d2e-87ce-c5099b76d43f",{"id":1950,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":1951,"copyright":18,"fieldtype":19,"meta_data":1952,"is_external_url":198},207774751179881,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2176\u002Fd4ea4f9c94\u002Fhydra-cost-speed.png",{"size":1823},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1954},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":1956},{},"The same learning cuts waste: cost per completed run and run duration fall round over round. Lower is better.",{"_uid":1959,"anchor":18,"content":1960,"eyebrow":18,"component":1515},"6cbc50c2-6492-4c07-af49-47960a16cdac",{"type":28,"attrs":1961,"content":1962},{"backgroundColor":17},[1963],{"type":31,"attrs":1964,"content":1965},{"textAlign":17},[1966],{"text":1967,"type":36},"The gains came from generalization rather than recall. A finding that only fit the scenario it came from had nowhere else to apply. The evidence that this transfers is production itself: since Hydra was introduced, operational completion across the procedures Pointer runs live has held above 99%.",{"_uid":1969,"anchor":1970,"content":1971,"eyebrow":18,"component":1515},"5c819cce-a5b0-4e35-a1d1-503e273c2838","Inside the improvement loop",{"type":28,"attrs":1972,"content":1973},{"backgroundColor":17},[1974,1978,1983],{"type":911,"attrs":1975,"content":1976},{"level":1468,"textAlign":17},[1977],{"text":1970,"type":36},{"type":31,"attrs":1979,"content":1980},{"textAlign":17},[1981],{"text":1982,"type":36},"Hydra reads completed runs. It sees the full trajectory, the tool calls, the intermediate outputs, and how the run ended, and it looks for patterns that hold beyond a single run, analyzing runs individually and in batches. Everything it extracts is recorded as a finding, and no finding influences future work until it is validated. Every finding stays scoped to its procedure, carries the evidence it came from, and can be walked back. Findings take one of three forms.",{"type":1984,"attrs":1985,"content":1986},"ordered_list",{"order":1641},[1987,1998,2009],{"type":1583,"content":1988},[1989],{"type":31,"attrs":1990,"content":1991},{"textAlign":17},[1992,1996],{"text":1993,"type":36,"marks":1994},"Learnings",[1995],{"type":183},{"text":1997,"type":36}," change how the system goes about the work without rewriting the procedure. In one procedure, the claims portal URL carried an internal identifier that differed from the official claim number displayed on the page, and the runtime kept trusting the URL. Hydra extracted a learning to use the displayed number. That learning is active today, with 55 successful applications out of 56. These are the closest thing to traditional memory. The difference lies in how much authority they carry. The effect of each active learning stays interpretable rather than dissolving into a bundle of simultaneous memories.",{"type":1583,"content":1999},[2000],{"type":31,"attrs":2001,"content":2002},{"textAlign":17},[2003,2007],{"text":2004,"type":36,"marks":2005},"Rule suggestions",[2006],{"type":183},{"text":2008,"type":36}," are material changes to the procedure itself. If a step is underspecified, Hydra can propose the change but it cannot silently make it policy. In another procedure, six consecutive runs attempted extraction without a readable source document. Hydra proposed an input gate: confirm a parseable document exists before invoking the model, and fail with a clear input error otherwise. An operator accepted the rule, and all six work items completed on the next pass. This is a design-time decision about the standard the work runs against, made once, off the execution path. It is not a person sitting in the loop on production runs.",{"type":1583,"content":2010},[2011],{"type":31,"attrs":2012,"content":2013},{"textAlign":17},[2014,2018],{"text":2015,"type":36,"marks":2016},"Issues",[2017],{"type":183},{"text":2019,"type":36}," capture problems the runtime cannot safely self-correct such as missing capabilities, broken integrations, infrastructure failures, or tooling defects. These are surfaced to the Pointer team for repair instead of being disguised as a prompt lesson.",{"_uid":2021,"image":2022,"layout":1828,"caption":2034,"component":1830},"2df5a3d3-3856-4bff-b2e7-2898df0f202f",[2023],{"_uid":2024,"image":2025,"video":2030,"lottie":2032,"component":203},"d7fef24b-1daa-4191-bd29-67f178d619c7",{"id":2026,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2027,"copyright":18,"fieldtype":19,"meta_data":2028,"is_external_url":198},207774347854941,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F4084x3168\u002Fab5c74d87b\u002Fhydra-architecture.png",{"size":2029},"4084x3168",{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2031},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2033},{},"Hydra's finding loop: evidence from completed runs becomes trajectory analysis, then a rule suggestion, learning, or issue, each gated before it can reach future runs.",{"_uid":2036,"anchor":18,"content":2037,"eyebrow":18,"component":1515},"a3b41bfa-75b4-4b1c-9afe-c0203e1ab821",{"type":28,"attrs":2038,"content":2039},{"backgroundColor":17},[2040,2045],{"type":31,"attrs":2041,"content":2042},{"textAlign":17},[2043],{"text":2044,"type":36},"Most of these surface early when we work with a customer. During the testing and simulation that precede any production deployment, a procedure runs against the adversarial scenarios first, and that is where the underspecified steps get caught and written in. The bulk of the rule suggestions land before the work is ever trusted with live operations, which is what compresses the time between standing a procedure up and running it unattended. The system keeps finding refinements in production, but it does not start from zero there.",{"type":31,"attrs":2046,"content":2047},{"textAlign":17},[2048],{"text":2049,"type":36},"Reaching the right answer and proving it is right are two different problems. Hydra focuses on the first. The second is a separate system we will get into another time.",{"_uid":2051,"anchor":2052,"content":2053,"eyebrow":18,"component":1515},"116b3f0d-a917-4e3d-b3f6-e0c3bf47b027","What we learned",{"type":28,"attrs":2054,"content":2055},{"backgroundColor":17},[2056,2060,2065,2101,2106,2111],{"type":911,"attrs":2057,"content":2058},{"level":1468,"textAlign":17},[2059],{"text":2052,"type":36},{"type":31,"attrs":2061,"content":2062},{"textAlign":17},[2063],{"text":2064,"type":36},"Implementing Hydra strengthened several hypotheses we held and weakened one the field still takes for granted.",{"type":1984,"attrs":2066,"content":2067},{"order":1641},[2068,2079,2090],{"type":1583,"content":2069},[2070],{"type":31,"attrs":2071,"content":2072},{"textAlign":17},[2073,2077],{"text":2074,"type":36,"marks":2075},"Repeated execution evidence can reveal reusable improvements.",[2076],{"type":183},{"text":2078,"type":36}," The operational recoveries and correctness gains along with cost and duration improvements show that both failed and successful runs contain information worth preserving.",{"type":1583,"content":2080},[2081],{"type":31,"attrs":2082,"content":2083},{"textAlign":17},[2084,2088],{"text":2085,"type":36,"marks":2086},"Imperfect instructions are a normal operating condition, not a defect to design around.",[2087],{"type":183},{"text":2089,"type":36}," A system that only improves under exhaustive instructions is solving the friendliest version of the problem. The procedures that gained the most were the ones whose SOPs left the most unsaid.",{"type":1583,"content":2091},[2092],{"type":31,"attrs":2093,"content":2094},{"textAlign":17},[2095,2099],{"text":2096,"type":36,"marks":2097},"Correctness needs several independent measures.",[2098],{"type":183},{"text":2100,"type":36}," Completion alone is not enough. Decisions, numerical values, required fields, and structure each fail on their own, and a single score hides all of it.",{"type":31,"attrs":2102,"content":2103},{"textAlign":17},[2104],{"text":2105,"type":36},"The one we came in believing and left doubting is that more memory makes a better agent. It does not. The reason is mechanical - retrieval runs on similarity, so a memory that is merely topically related to the task at hand gets surfaced even when it is stale or beside the point, and once it is in the context window the model tends to lean on it and fold it into the answer with confidence.",{"type":31,"attrs":2107,"content":2108},{"textAlign":17},[2109],{"text":2110,"type":36},"You see this in general-purpose assistants constantly. You are asking about one thing, and the model reaches into its saved memory and drops in a note from some unrelated past conversation. It surfaces because the retrieval step decided it was close enough, and the model has no reliable way to tell that it does not belong here. In a chat that is a harmless non sequitur you scroll past.",{"type":31,"attrs":2112,"content":2113},{"textAlign":17},[2114],{"text":2115,"type":36},"In a financial close, where a stray note can tilt a decision on an invoice it was never meant to touch, the same behavior is drift. Memory without scope and validation compounds into bias.",{"_uid":2117,"anchor":2118,"content":2119,"eyebrow":18,"component":1515},"3c24935d-e558-41f8-b805-1c5b360579a7","The compounding advantage that stays with you",{"type":28,"attrs":2120,"content":2121},{"backgroundColor":17},[2122,2126,2131,2136,2141,2146,2151,2156,2172,2177,2182],{"type":911,"attrs":2123,"content":2124},{"level":1468,"textAlign":17},[2125],{"text":2118,"type":36},{"type":31,"attrs":2127,"content":2128},{"textAlign":17},[2129],{"text":2130,"type":36},"The reason someone who has run reconciliation for ten years is so good at it is because they have simply seen it all. They know which vendor invoices never match on the first pass, which accounts always need a manual adjustment, which exceptions are real and which are noise. That knowledge is what lets them run the close without anyone checking behind them, and almost none of it is written down.",{"type":31,"attrs":2132,"content":2133},{"textAlign":17},[2134],{"text":2135,"type":36},"That is the problem underneath everything mentioned here. The expertise that makes operational work run lives in a few heads, and it leaves with them. It is also the honest worry about handing the work to AI.",{"type":31,"attrs":2137,"content":2138},{"textAlign":17},[2139],{"text":2140,"type":36},"A general model arrives knowing nothing about your particular process, and there is no manual to give it.",{"type":31,"attrs":2142,"content":2143},{"textAlign":17},[2144],{"text":2145,"type":36},"Hydra is how the system earns that competence without spending ten years on the job. Every run leaves evidence, and Hydra converts it into the judgment the work actually requires, validated and written down for the first time.",{"type":31,"attrs":2147,"content":2148},{"textAlign":17},[2149],{"text":2150,"type":36},"Most of that earning happens before production, in the simulation phase described above, so the system goes live already knowing the edge cases a new hire would take months to encounter.",{"type":31,"attrs":2152,"content":2153},{"textAlign":17},[2154],{"text":2155,"type":36},"That loop is what makes an enterprise deployment possible at all:",{"type":1580,"content":2157},[2158,2165],{"type":1583,"content":2159},[2160],{"type":31,"attrs":2161,"content":2162},{"textAlign":17},[2163],{"text":2164,"type":36},"No process is fully specified on day one, and a system that cannot close its own gaps turns every unwritten step into an exception routed to a person forever, so the team it was meant to free never leaves the loop.",{"type":1583,"content":2166},[2167],{"type":31,"attrs":2168,"content":2169},{"textAlign":17},[2170],{"text":2171,"type":36},"No enterprise can hand work to a system that rewrites its own behavior freely, and Hydra was built under that constraint.",{"type":31,"attrs":2173,"content":2174},{"textAlign":17},[2175],{"text":2176,"type":36},"What the company gets is competence that compounds while staying visible. Each run makes the system better at your work specifically, and that accumulated judgment lives in your deployment rather than in any single person or any model underneath. A better base model raises the floor for everyone, but the record of how your work is done right is yours alone.",{"type":31,"attrs":2178,"content":2179},{"textAlign":17},[2180],{"text":2181,"type":36},"This is what lets Pointer run multi-day reconciliations, three-way matches, and threshold calculations inside some of the most regulated companies in the world without a person checking behind every step.",{"type":31,"attrs":2183,"content":2184},{"textAlign":17},[2185,2187,2193],{"text":2186,"type":36},"If you’re excited by self-improving AI systems for work that has to be right, ",{"text":2188,"type":36,"marks":2189},"we’re hiring across all functions",[2190],{"type":256,"attrs":2191},{"href":2192,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fpointer.ai\u002Fcareers",{"text":2194,"type":36},".",{"id":2196,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2197,"copyright":18,"fieldtype":19,"meta_data":2198,"is_external_url":198},207771218658348,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F7320x4122\u002F027bcc2f0d\u002Fhydra-hero.png",{"size":2199},"7320x4122",{"type":28,"attrs":2201,"content":2202},{"backgroundColor":17},[2203],{"type":31,"attrs":2204,"content":2205},{"textAlign":17},[2206],{"text":128,"type":36,"marks":2207},[2208],{"type":183},"engineering",{"name":2211,"created_at":2212,"published_at":2213,"updated_at":2214,"id":2215,"uuid":2216,"content":2217,"slug":2225,"full_slug":2226,"sort_by_date":17,"position":413,"tag_list":2227,"is_startpage":198,"parent_id":2228,"meta_data":17,"group_id":2229,"first_published_at":2213,"release_id":17,"lang":418,"path":17,"alternates":2230,"default_full_slug":17,"translated_slugs":17,"_stopResolving":109},"Aryaman Shandilya","2026-08-11T02:51:08.035Z","2026-08-11T02:54:18.591Z","2026-08-11T02:54:18.608Z",207778070685160,"18a0fdda-b439-4e5a-8bb2-bd2957d99c75",{"Name":2211,"_uid":2218,"Headshot":2219,"component":2224},"afae72ca-36e6-472a-b3fa-e197557637b9",{"id":2220,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2221,"copyright":18,"fieldtype":19,"meta_data":2222,"is_external_url":198},207778220766415,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F512x512\u002Fabbd870355\u002Faryaman.png",{"size":2223},"512x512","author","aryaman-shandilya","blog\u002Fauthors\u002Faryaman-shandilya",[],165966507666448,"9550de2d-59c7-4276-bdce-13c53aceb250",[],{"type":28,"attrs":2232,"content":2233},{"backgroundColor":17},[2234],{"type":31,"attrs":2235,"content":2236},{"textAlign":17},[2237],{"text":1457,"type":36},"article","2026-08-10 00:00",[],-80,[],165966492818446,"8d389ac3-853c-47a8-bc00-9b853a87cb3e","2026-08-11T02:54:06.879Z",[],[2248,2251,2254],{"id":2249,"name":2250,"value":2209,"dimension_value":17},165966787497574,"Engineering",{"id":2252,"name":2253,"value":21,"dimension_value":17},165966789889639,"Company",{"id":2255,"name":2256,"value":2257,"dimension_value":17},165966791642728,"Industry","industry",{"stories":2259,"total":1641},[2260],{"name":106,"created_at":2261,"published_at":2262,"updated_at":2263,"id":107,"uuid":103,"content":2264,"slug":108,"full_slug":104,"sort_by_date":17,"position":3455,"tag_list":3456,"is_startpage":198,"parent_id":2243,"meta_data":17,"group_id":3457,"first_published_at":3458,"release_id":17,"lang":418,"path":17,"alternates":3459,"default_full_slug":17,"translated_slugs":17},"2026-05-21T21:19:48.858Z","2026-06-25T19:50:15.198Z","2026-06-25T19:50:15.225Z",{"seo":2265,"_uid":1458,"body":2273,"image":3419,"title":3421,"topic":2209,"author":3429,"excerpt":3446,"component":2238,"publishedOn":3453,"hide_on_index":198,"related_articles":3454},[2266],{"_uid":1450,"image":2267,"title":2271,"component":436,"description":2272},{"id":2268,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2269,"copyright":18,"fieldtype":19,"meta_data":2270,"is_external_url":198},180200290135589,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F7352x4136\u002Faed0b956a6\u002Fosworld-hero.png",{},"Pointer - A new state of the art for computer use","We just hit a new state of the art on OSWorld. This is how we built the agent, what surprised us along the way, and why benchmark performance is only the beginning of an exciting new primitive.",[2274,2338,2352,2391,2437,2449,2657,2669,2746,2760,2783,2795,2809,2833,2847,2865,2879,2917,2931,2988,3002,3058,3256,3297],{"_uid":1461,"anchor":2275,"content":2276,"eyebrow":18,"component":1515},"Overview",{"type":28,"content":2277},[2278,2282,2287,2292,2303,2308,2328,2333],{"type":911,"attrs":2279,"content":2280},{"level":1468,"textAlign":17},[2281],{"text":2275,"type":36},{"type":31,"attrs":2283,"content":2284},{"textAlign":17},[2285],{"text":2286,"type":36},"Today we’re sharing a new state of the art for computer use agents.",{"type":31,"attrs":2288,"content":2289},{"textAlign":17},[2290],{"text":2291,"type":36},"We hold the two highest verified scores on the OSWorld leaderboard to date: 83.6% with Claude Opus 4.7, and 81.5% with Claude Sonnet 4.6. The human baseline on this benchmark is 72.4%.",{"type":31,"attrs":2293,"content":2294},{"textAlign":17},[2295,2301],{"text":2296,"type":36,"marks":2297},"OSWorld",[2298],{"type":256,"attrs":2299},{"href":2300,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fos-world.github.io\u002F",{"text":2302,"type":36}," is the standard benchmark for computer use agents – AI that operates a real computer the same way a human does. It’s how the field measures progress on this problem, and the leaderboard is a mix of frontier labs and research groups.",{"type":31,"attrs":2304,"content":2305},{"textAlign":17},[2306],{"text":2307,"type":36},"This result matters to the people we build for. It’s a proxy for a simple commitment: the agent operating their systems is the best one available to do it.",{"type":31,"attrs":2309,"content":2310},{"textAlign":17},[2311,2313,2319,2321,2326],{"text":2312,"type":36},"We’re releasing the full system as ",{"text":2314,"type":36,"marks":2315},"open source",[2316],{"type":256,"attrs":2317},{"href":2318,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fgithub.com\u002FPointer-so\u002FOSWorld\u002Ftree\u002Fmain",{"text":2320,"type":36},", and the result is on the ",{"text":2322,"type":36,"marks":2323},"public leaderboard",[2324],{"type":256,"attrs":2325},{"href":2300,"uuid":17,"anchor":17,"target":17,"linktype":172},{"text":2327,"type":36},". What follows is how we did it: the architecture, what worked, what surprised us, and what a benchmark like this can and can’t tell us about computer use in the real world.",{"type":31,"attrs":2329,"content":2330},{"textAlign":17},[2331],{"text":2332,"type":36},"The two scores come from the same system. We ran it twice and changed only the model doing the work. This swap revealed interesting insights about model behavior and routing that we return to throughout this post.",{"type":31,"attrs":2334,"content":2335},{"textAlign":17},[2336],{"text":2337,"type":36},"Thanks to the team behind OSWorld for building the benchmark, and to the researchers who came before us for the work we learned from.",{"_uid":2339,"image":2340,"layout":1828,"caption":2351,"component":1830},"949b1558-8ad7-4543-8a7b-b82a9e9394df",[2341],{"_uid":2342,"image":2343,"video":2347,"lottie":2349,"component":203},"10f9a0ed-5829-42d2-8d69-3f542116b07e",{"id":2344,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2345,"copyright":18,"fieldtype":19,"meta_data":2346,"is_external_url":198},180562157985897,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2832\u002F2898a0f668\u002Fresults-overview.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2348},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2350},{},"OSWorld leaderboard (includes agentic frameworks, specialized models, and general models) showing our top two verified scores as of May 26, 2026. Both the Opus 4.7 and Sonnet 4.6 runs outperformed the previous best agent and the human baseline of 72.4%.",{"_uid":1517,"anchor":2353,"content":2354,"eyebrow":18,"component":1515},"What is computer use",{"type":28,"content":2355},[2356,2361,2366,2371,2376,2381,2386],{"type":911,"attrs":2357,"content":2358},{"level":1468,"textAlign":17},[2359],{"text":2360,"type":36},"What is computer use?",{"type":31,"attrs":2362,"content":2363},{"textAlign":17},[2364],{"text":2365,"type":36},"OSWorld runs an agent in a real desktop environment – a real operating system, applications, files, browsers, and settings – and asks it to complete tasks in the same way a person would: clicking, typing, and navigating.",{"type":31,"attrs":2367,"content":2368},{"textAlign":17},[2369],{"text":2370,"type":36},"It spans more than 360 tasks across web browsing (Chrome), coding (VS Code), photo editing (GIMP), document and spreadsheet work (LibreOffice), media playback (VLC), email (Thunderbird), and the operating system itself. Many tasks require traversing multiple applications to reach an outcome. Some are deliberately impossible, meaning the agent is also scored on its ability to recognize a dead end.",{"type":31,"attrs":2372,"content":2373},{"textAlign":17},[2374],{"text":2375,"type":36},"This is what makes it a meaningful test. Most software was built for people, not programs, and the interface it exposes to the world is the screen. APIs only ever cover a narrow, intentionally chosen slice of what an application can do; the screen exposes all of it.",{"type":31,"attrs":2377,"content":2378},{"textAlign":17},[2379],{"text":2380,"type":36},"An agent that can operate the screen isn’t limited to the integrations someone thought to build in advance. It can, in principle, do anything a person sitting at the computer can do.",{"type":31,"attrs":2382,"content":2383},{"textAlign":17},[2384],{"text":2385,"type":36},"This is also what separates computer use from the automation that came before it. Robotic process automation (RPA) follows a fixed script: click here, then here, then type this. A computer use agent is the inverse. You give it the goal, and it figures out how to get there from what’s actually in front of it – which is why it survives the messiness that breaks a script. That distinction matters most in real environments, where the systems are old, inconsistent, and were never designed to be automated in the first place.",{"type":31,"attrs":2387,"content":2388},{"textAlign":17},[2389],{"text":2390,"type":36},"The pace here has been quick. A year ago the best results on OSWorld hovered around 20%. Today the frontier has cleared the human baseline of 72.4%, with the strongest individual models scoring in the high seventies and full systems now past 83%.",{"_uid":1560,"anchor":2392,"content":2393,"eyebrow":18,"component":1515},"The architecture",{"type":28,"content":2394},[2395,2399,2404,2409,2414],{"type":911,"attrs":2396,"content":2397},{"level":1468,"textAlign":17},[2398],{"text":2392,"type":36},{"type":31,"attrs":2400,"content":2401},{"textAlign":17},[2402],{"text":2403,"type":36},"The most important decision we made was to keep the system simple.",{"type":31,"attrs":2405,"content":2406},{"textAlign":17},[2407],{"text":2408,"type":36},"Previous agentic systems required elaborate scaffolding, such as fine-tuning grounding models to locate screen elements or building long chains of special-case routing logic. We started down some of that road and kept deleting it. Frontier models have improved at these base capabilities. What remains is a lightweight task controller and three specialized agents, each with a single responsibility.",{"type":31,"attrs":2410,"content":2411},{"textAlign":17},[2412],{"text":2413,"type":36},"As shown in the architecture diagram, the task controller acts as a central orchestrator. It treats the agents as method tool calls and runs them in a strict sequence:",{"type":1580,"content":2415},[2416,2423,2430],{"type":1583,"content":2417},[2418],{"type":31,"attrs":2419,"content":2420},{"textAlign":17},[2421],{"text":2422,"type":36},"A feasibility gate decides whether the task is possible at all before any work begins.",{"type":1583,"content":2424},[2425],{"type":31,"attrs":2426,"content":2427},{"textAlign":17},[2428],{"text":2429,"type":36},"A planner turns a possible task into a short list of target milestones.",{"type":1583,"content":2431},[2432],{"type":31,"attrs":2433,"content":2434},{"textAlign":17},[2435],{"text":2436,"type":36},"An executor runs in a continuous inner loop to carry them out.",{"_uid":1815,"image":2438,"layout":1828,"caption":2448,"component":1830},[2439],{"_uid":1818,"image":2440,"video":2444,"lottie":2446,"component":203},{"id":2441,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2442,"copyright":18,"fieldtype":19,"meta_data":2443,"is_external_url":198},180558776705049,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F4704x2452\u002Ffead358cf4\u002Fagent-architecture.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2445},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2447},{},"System architecture flow. A central task controller coordinates the sequence, treating each specialized agent as a tool call. Only the executor interacts directly with the virtual machine to change the state of the desktop.",{"_uid":1832,"anchor":18,"content":2450,"eyebrow":18,"component":1515},{"type":28,"content":2451},[2452,2457,2568,2573,2578,2584,2589,2594,2603,2608,2626,2631,2636,2647,2652],{"type":31,"attrs":2453,"content":2454},{"textAlign":17},[2455],{"text":2456,"type":36},"Each runs on the model best suited to its job:",{"type":1633,"content":2458},[2459,2491,2517,2542],{"type":1636,"content":2460},[2461,2471,2481],{"type":1639,"attrs":2462,"content":2463},{"colspan":1641,"rowspan":1641,"colwidth":17},[2464],{"type":31,"attrs":2465,"content":2466},{"textAlign":17},[2467],{"text":2468,"type":36,"marks":2469},"Agent",[2470],{"type":183},{"type":1639,"attrs":2472,"content":2473},{"colspan":1641,"rowspan":1641,"colwidth":17},[2474],{"type":31,"attrs":2475,"content":2476},{"textAlign":17},[2477],{"text":2478,"type":36,"marks":2479},"Model",[2480],{"type":183},{"type":1639,"attrs":2482,"content":2483},{"colspan":1641,"rowspan":1641,"colwidth":17},[2484],{"type":31,"attrs":2485,"content":2486},{"textAlign":17},[2487],{"text":2488,"type":36,"marks":2489},"Role",[2490],{"type":183},{"type":1636,"content":2492},[2493,2501,2509],{"type":1675,"attrs":2494,"content":2495},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2496],{"type":31,"attrs":2497,"content":2498},{"textAlign":17},[2499],{"text":2500,"type":36},"Feasibility gate",{"type":1675,"attrs":2502,"content":2503},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2504],{"type":31,"attrs":2505,"content":2506},{"textAlign":17},[2507],{"text":2508,"type":36},"Claude Sonnet 4.6",{"type":1675,"attrs":2510,"content":2511},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2512],{"type":31,"attrs":2513,"content":2514},{"textAlign":17},[2515],{"text":2516,"type":36},"Decide whether the task is achievable",{"type":1636,"content":2518},[2519,2527,2534],{"type":1675,"attrs":2520,"content":2521},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2522],{"type":31,"attrs":2523,"content":2524},{"textAlign":17},[2525],{"text":2526,"type":36},"Planner",{"type":1675,"attrs":2528,"content":2529},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2530],{"type":31,"attrs":2531,"content":2532},{"textAlign":17},[2533],{"text":2508,"type":36},{"type":1675,"attrs":2535,"content":2536},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2537],{"type":31,"attrs":2538,"content":2539},{"textAlign":17},[2540],{"text":2541,"type":36},"Break the goal into milestones",{"type":1636,"content":2543},[2544,2552,2560],{"type":1675,"attrs":2545,"content":2546},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2547],{"type":31,"attrs":2548,"content":2549},{"textAlign":17},[2550],{"text":2551,"type":36},"Executor",{"type":1675,"attrs":2553,"content":2554},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2555],{"type":31,"attrs":2556,"content":2557},{"textAlign":17},[2558],{"text":2559,"type":36},"Claude Opus 4.7",{"type":1675,"attrs":2561,"content":2562},{"colspan":1641,"rowspan":1641,"colwidth":17,"backgroundColor":17},[2563],{"type":31,"attrs":2564,"content":2565},{"textAlign":17},[2566],{"text":2567,"type":36},"Do the work – the only agent that changes the machine’s state",{"type":31,"attrs":2569,"content":2570},{"textAlign":17},[2571],{"text":2572,"type":36},"The table shows the configuration for our highest-scoring run. The executor is the one we swap (Opus 4.7 for our highest score, Sonnet 4.6 for the second) while the gate and planner stayed on Sonnet 4.6 throughout.",{"type":31,"attrs":2574,"content":2575},{"textAlign":17},[2576],{"text":2577,"type":36},"A smaller model, Claude Haiku 4.5, handles context compaction in the background. It summarizes older steps when a task runs long enough to strain the primary model’s context window.",{"type":911,"attrs":2579,"content":2581},{"level":2580,"textAlign":17},5,[2582],{"text":2583,"type":36},"The feasibility gate decides what not to attempt",{"type":31,"attrs":2585,"content":2586},{"textAlign":17},[2587],{"text":2588,"type":36},"Models suffer from unbounded optimism. Their default behavior is to attempt a task even if the request is absurd.",{"type":31,"attrs":2590,"content":2591},{"textAlign":17},[2592],{"text":2593,"type":36},"Roughly 1 in 13 tasks on the OSWorld benchmark is deliberately impossible. An agent without a filter will often proceed anyway and fabricate a plausible-looking end state. In real-world applications, this type of hallucination is dangerous because a false positive is much worse than simply asking a human for help.",{"type":31,"attrs":2595,"content":2596},{"textAlign":17},[2597,2599],{"text":2598,"type":36},"Before any work begins, the feasibility gate asks a question many agents skip: ",{"text":2600,"type":36,"marks":2601},"can this task be done at all?",[2602],{"type":1481},{"type":31,"attrs":2604,"content":2605},{"textAlign":17},[2606],{"text":2607,"type":36},"It uses read-only tools like shell commands and Python scripts to probe the live environment. It checks installed software and connected hardware, then evaluates the prompt against concrete criteria. Our baseline rule for this agent is strict – if a task requires an extreme workaround, it is infeasible. Most real software does not require sleight of hand for basic functions.",{"type":31,"attrs":2609,"content":2610},{"textAlign":17},[2611,2613,2618,2620,2624],{"text":2612,"type":36},"A concrete case: one task asks to “automatically adjust the brightness and contrast of this video to match my room’s lighting.” The gate confirmed the VLC media player was installed. It then scanned the machine for ambient light sensors by reading paths like ",{"text":2614,"type":36,"marks":2615},"\u002Fdev\u002Fvideo*",[2616],{"type":2617},"code",{"text":2619,"type":36}," and ",{"text":2621,"type":36,"marks":2622},"\u002Fsys\u002Fclass\u002Fbacklight\u002F",[2623],{"type":2617},{"text":2625,"type":36},". Its verdict had two layers: the machine has no sensor that could measure room lighting, and even if it did, VLC has no feature that could act on the reading. The task was doubly infeasible, so the agent correctly declined it.",{"type":31,"attrs":2627,"content":2628},{"textAlign":17},[2629],{"text":2630,"type":36},"The agent rejected the prompt based on evidence it actively gathered rather than relying on learned sycophancy. Teaching a model this kind of operational humility, the ability to recognize an impossible request and stop before inventing a result, is just as important as teaching it to execute tasks.",{"type":911,"attrs":2632,"content":2633},{"level":2580,"textAlign":17},[2634],{"text":2635,"type":36},"The planner describes states, not steps",{"type":31,"attrs":2637,"content":2638},{"textAlign":17},[2639,2641,2645],{"text":2640,"type":36},"The planner’s output is a list of milestones, and each milestone describes a ",{"text":2642,"type":36,"marks":2643},"state the world should be in",[2644],{"type":1481},{"text":2646,"type":36},", not the recipe for getting there.",{"type":31,"attrs":2648,"content":2649},{"textAlign":17},[2650],{"text":2651,"type":36},"For example, a milestone might specify “the spreadsheet is sorted by date and saved” rather than “click the Data menu, then Sort, then OK.” The executor decides the how, while the planner only specifies the where.",{"type":31,"attrs":2653,"content":2654},{"textAlign":17},[2655],{"text":2656,"type":36},"This sounds like a small distinction but isn’t. The moment you hand an agent a sequence of clicks, you’ve made it brittle. An unexpected dialog box or a slightly moved menu becomes fatal. A milestone aimed at a state survives all of that, because the executor is free to find another route. It can also create sub-milestones and revise the plan mid-task when it finds evidence the original approach won’t work.",{"_uid":1945,"image":2658,"layout":1828,"caption":2668,"component":1830},[2659],{"_uid":1948,"image":2660,"video":2664,"lottie":2666,"component":203},{"id":2661,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2662,"copyright":18,"fieldtype":19,"meta_data":2663,"is_external_url":198},180559145725978,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x4308\u002F5c42049026\u002Fplanner-correction.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2665},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2667},{},"The executor revising a plan mid-task. Upon discovering that the data totals sit in column I rather than the planned column J, it rewrites the affected milestones and continues. A fixed sequence of didactic instructions would have failed here.",{"_uid":1959,"anchor":18,"content":2670,"eyebrow":18,"component":1515},{"type":28,"content":2671},[2672,2677,2682,2687,2696,2701,2731,2736,2741],{"type":911,"attrs":2673,"content":2674},{"level":2580,"textAlign":17},[2675],{"text":2676,"type":36},"One executor with every tool",{"type":31,"attrs":2678,"content":2679},{"textAlign":17},[2680],{"text":2681,"type":36},"Early on in our experimentation, the executor was a router that handed tasks to specialist sub-agents – one for code, one for the GUI, and another for the browser. We ultimately collapsed this into a single executor with every tool available and let the model decide what to reach for.",{"type":31,"attrs":2683,"content":2684},{"textAlign":17},[2685],{"text":2686,"type":36},"Historically, large tool schemas caused models to lose focus, making sub-agents necessary. Today’s frontier can manage extensive toolsets without degrading performance. More importantly, abandoning sub-agents removes artificial boundaries.",{"type":31,"attrs":2688,"content":2689},{"textAlign":17},[2690,2694],{"text":2691,"type":36,"marks":2692},"Real computer use is highly cross-modal.",[2693],{"type":183},{"text":2695,"type":36}," If a “code agent” runs a script and a “GUI agent” checks the output, the handoff inevitably loses nuances. A unified executor maintains perfect state continuity. It can write a script, read the terminal, and immediately use the mouse without summarizing its state for someone else.",{"type":31,"attrs":2697,"content":2698},{"textAlign":17},[2699],{"text":2700,"type":36},"Its tools fall into a few groups:",{"type":1580,"content":2702},[2703,2710,2717,2724],{"type":1583,"content":2704},[2705],{"type":31,"attrs":2706,"content":2707},{"textAlign":17},[2708],{"text":2709,"type":36},"GUI – click, type, scroll, drag, hotkeys, window management. The basic vocabulary of a person at a keyboard and mouse.",{"type":1583,"content":2711},[2712],{"type":31,"attrs":2713,"content":2714},{"textAlign":17},[2715],{"text":2716,"type":36},"Code – direct Python and Bash execution for installing packages, reading and writing files, and anything more naturally done in a few lines than by hand.",{"type":1583,"content":2718},[2719],{"type":31,"attrs":2720,"content":2721},{"textAlign":17},[2722],{"text":2723,"type":36},"Browser – direct access to Chrome through the DevTools Protocol for reading page state cleanly.",{"type":1583,"content":2725},[2726],{"type":31,"attrs":2727,"content":2728},{"textAlign":17},[2729],{"text":2730,"type":36},"Long-running work – a background-execution tool for commands that take longer than the system’s two-minute limit.",{"type":31,"attrs":2732,"content":2733},{"textAlign":17},[2734],{"text":2735,"type":36},"Two of these tools work around strict environmental constraints.",{"type":31,"attrs":2737,"content":2738},{"textAlign":17},[2739],{"text":2740,"type":36},"The browser tool must connect to Chrome’s debugging port, which Chrome binds exclusively to the local machine for security. Because our agent runs outside the VM, the executor ships small Python snippets into the machine. These snippets talk to Chrome locally and return only the requested page state.",{"type":31,"attrs":2742,"content":2743},{"textAlign":17},[2744],{"text":2745,"type":36},"The long-running tool exists because the environment forcefully kills any command that exceeds 120 seconds. This limit caused some of our strangest failures early on. It is fine for most UI actions but fatal for installing large packages or running slow scripts. Worse, the agent had no way to recover from being abruptly cut off. Instead of holding a connection open and hoping it survives, the executor uses the background tool to launch the command detached. It receives a process handle immediately and polls for completion on its own schedule.",{"_uid":2747,"image":2748,"layout":1828,"caption":2759,"component":1830},"c88c552d-35d6-468f-b149-dad9e9d3f4bd",[2749],{"_uid":2750,"image":2751,"video":2755,"lottie":2757,"component":203},"ca387491-747e-4740-abc4-66d3f5d112a1",{"id":2752,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2753,"copyright":18,"fieldtype":19,"meta_data":2754,"is_external_url":198},180560539533394,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2064\u002F45ede5dfaf\u002Flong-running-work.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2756},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2758},{},"The executor recovering from the two-minute limit. A run_bash job is killed at 2:00; the executor recognizes the timeout and re-runs the work detached with run_background, polling until it completes rather than retrying into the same wall.",{"_uid":1969,"anchor":18,"content":2761,"eyebrow":18,"component":1515},{"type":28,"content":2762},[2763,2768,2773,2778],{"type":911,"attrs":2764,"content":2765},{"level":2580,"textAlign":17},[2766],{"text":2767,"type":36},"GUI first",{"type":31,"attrs":2769,"content":2770},{"textAlign":17},[2771],{"text":2772,"type":36},"Models naturally default to writing code. Their training data is heavily saturated with programming languages and command-line utilities. As a result, when faced with a task like updating a spreadsheet, a frontier model will instinctively try to use a library to modify the underlying file directly. Code is fast, deterministic, and highly rewarded during post-training.",{"type":31,"attrs":2774,"content":2775},{"textAlign":17},[2776],{"text":2777,"type":36},"We explicitly instructed our executor to prioritize GUIs over these programmatic workarounds. We did this partly because OSWorld evaluates the actual desktop environment.",{"type":31,"attrs":2779,"content":2780},{"textAlign":17},[2781],{"text":2782,"type":36},"A spreadsheet generated entirely by a Python library might look correct to a human reader, but it often lacks the specific XML metadata or internal object structures that the actual application creates. More importantly, forcing the agent to use the screen aligns with the core purpose of the benchmark. The goal is to measure how well an AI operates a computer the way a human does.",{"_uid":2021,"image":2784,"layout":1828,"caption":2794,"component":1830},[2785],{"_uid":2024,"image":2786,"video":2790,"lottie":2792,"component":203},{"id":2787,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2788,"copyright":18,"fieldtype":19,"meta_data":2789,"is_external_url":198},180560721403993,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x608\u002F04e1b17f47\u002Fgui-vs-code-percent-distribution.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2791},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2793},{},"Distribution of state-changing tool calls across our Opus 4.7 run, each shown as a percentage of all state-changing calls. The agent relies heavily on UI actions like clicking and typing to execute tasks but frequently uses Bash and Python for intermediate data extraction. ",{"_uid":2036,"anchor":18,"content":2796,"eyebrow":18,"component":1515},{"type":28,"content":2797},[2798],{"type":31,"attrs":2799,"content":2800},{"textAlign":17},[2801,2803,2807],{"text":2802,"type":36},"This does not mean the agent abandons code entirely. As the distribution of tool calls shows, most trajectories blend both methods. A model will frequently drop into a terminal to write a quick Bash or Python script to locate a hidden file, parse a dense ",{"text":2804,"type":36,"marks":2805},".mbox",[2806],{"type":2617},{"text":2808,"type":36}," email archive, or extract an exact dollar amount from a PDF. Once the agent holds the parsed data in its context window, it switches back to the mouse and keyboard to complete the task inside the target application. This hybrid uses the model’s strong coding background for heavy data extraction while keeping its final actions grounded in realistic software operation.",{"_uid":2051,"anchor":2810,"content":2811,"eyebrow":18,"component":1515},"Results",{"type":28,"content":2812},[2813,2817,2822],{"type":911,"attrs":2814,"content":2815},{"level":1468,"textAlign":17},[2816],{"text":2810,"type":36},{"type":31,"attrs":2818,"content":2819},{"textAlign":17},[2820],{"text":2821,"type":36},"Our highest run (using Opus 4.7) scored 83.6% across the benchmark’s 361 tasks – 290 solved completely and 14 partially. The Sonnet 4.6 run scored 81.5%.",{"type":31,"attrs":2823,"content":2824},{"textAlign":17},[2825,2827,2831],{"text":2826,"type":36},"The OSWorld leaderboard is a mix of approaches. Some entries are base models reporting their own scores, some are full systems built around a model, some are specialized solely for the benchmark. Others report an average across runs or best across multiple rollouts (best-of-N, noted as ",{"text":2828,"type":36,"marks":2829},"bBoN",[2830],{"type":2617},{"text":2832,"type":36}," on the leaderboard). Ours are single runs, and they are the two highest recorded to date.",{"_uid":2834,"image":2835,"layout":1828,"caption":2846,"component":1830},"a870ef9a-7fc0-4af8-a07e-26707c8108bb",[2836],{"_uid":2837,"image":2838,"video":2842,"lottie":2844,"component":203},"92233f95-34d2-4f31-afa7-b3d788f37744",{"id":2839,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2840,"copyright":18,"fieldtype":19,"meta_data":2841,"is_external_url":198},180561113739357,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2176\u002F99c67d0838\u002Fresults-bar-chart.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2843},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2845},{},"OSWorld leaderboard (top five scores) showing our top two verified scores (in purple) as of May 26, 2026. Both the Opus 4.7 and Sonnet 4.6 runs clear the previous best. The human baseline is 72.4%.",{"_uid":2117,"anchor":18,"content":2848,"eyebrow":18,"component":1515},{"type":28,"content":2849},[2850,2855,2860],{"type":911,"attrs":2851,"content":2852},{"level":2580,"textAlign":17},[2853],{"text":2854,"type":36},"Where the agent is strong",{"type":31,"attrs":2856,"content":2857},{"textAlign":17},[2858],{"text":2859,"type":36},"The headline score is a single figure for more than 360 tasks across ten very different domains. The per-domain breakdown is more useful because it shows exactly where the agent is reliable and where it still slips.",{"type":31,"attrs":2861,"content":2862},{"textAlign":17},[2863],{"text":2864,"type":36},"The table below displays our best result in each domain across both runs (simulating a production system that routes tasks to the better-performing model). We compare this against the single best score anyone else has posted, as well as the mean of the top five competing agentic frameworks. The delta column compares our score to that mean, where a positive number indicates a lead.",{"_uid":2866,"image":2867,"layout":1828,"caption":2878,"component":1830},"f9afe707-0fd7-47eb-a3b3-766fc9881533",[2868],{"_uid":2869,"image":2870,"video":2874,"lottie":2876,"component":203},"071bcf1a-4353-4ef9-b1d0-105cff049e56",{"id":2871,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2872,"copyright":18,"fieldtype":19,"meta_data":2873,"is_external_url":198},180561317617762,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x3088\u002F03003caa5a\u002Fper-domain-comparison.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2875},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2877},{},"Our best score in each domain across both runs (Opus 4.7 and Sonnet 4.6), against the best result any agent has posted and its holder. The final column compares our score against the mean of the top five competing frameworks. We hold the top score in 7 of the 10 domains.",{"_uid":2880,"anchor":18,"content":2881,"eyebrow":18,"component":1515},"d3d65449-5d1f-41a9-ae1e-5f46abe1537b",{"type":28,"content":2882},[2883,2888,2893,2902,2907,2912],{"type":31,"attrs":2884,"content":2885},{"textAlign":17},[2886],{"text":2887,"type":36},"We hold or share the top spot in 7 of the 10 domains, and we beat the top-five mean in 9 of them. Notably, the two models do not lead the same domains. Opus takes most of them, but Sonnet is ahead on Impress, VS Code, and VLC. This validates treating the model as a swappable component rather than the entire system – different tasks demand different strengths.",{"type":31,"attrs":2889,"content":2890},{"textAlign":17},[2891],{"text":2892,"type":36},"The result we care about most is multi-apps. It is the largest category on the benchmark (93 tasks, roughly a quarter of the total) and the hardest, because the work spans several applications and requires the agent to carry state across them.",{"type":31,"attrs":2894,"content":2895},{"textAlign":17},[2896,2900],{"text":2897,"type":36,"marks":2898},"It is also the domain that looks most like real work.",[2899],{"type":183},{"text":2901,"type":36}," Almost nothing in a real enterprise happens inside a single window. Most systems struggle here. The field clusters in the low-to-mid sixties, while we scored 74.4%, several points clear of the next-best result.",{"type":31,"attrs":2903,"content":2904},{"textAlign":17},[2905],{"text":2906,"type":36},"The same pattern holds across standard productivity applications. Calc, Impress, and Writer all land above 91%. VS Code hits 95.7% and operating-system tasks reach 95.8%. These are the environments closest to the actual back-office work we build for.",{"type":911,"attrs":2908,"content":2909},{"level":2580,"textAlign":17},[2910],{"text":2911,"type":36},"The model is a component, not a system",{"type":31,"attrs":2913,"content":2914},{"textAlign":17},[2915],{"text":2916,"type":36},"Because our harness stayed fixed across our runs, the results isolate exactly what the scaffolding contributes to models with different baselines. We find this dynamic more interesting than the headline scores themselves, especially when factoring in cost.",{"_uid":2918,"image":2919,"layout":1828,"caption":2930,"component":1830},"3192b551-b165-4eb1-8eb4-b63b875ee48d",[2920],{"_uid":2921,"image":2922,"video":2926,"lottie":2928,"component":203},"0089268f-755b-408c-92a2-fb05dcb4b53f",{"id":2923,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2924,"copyright":18,"fieldtype":19,"meta_data":2925,"is_external_url":198},180584080339552,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x2176\u002F45bf106ce0\u002Fcost-vs-accuracy.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2927},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2929},{},"OSWorld accuracy against output-token cost per task. Base-model figures are sourced from each model’s system card, while the dotted lines connect them to their performance inside our harness. Both models gain accuracy: Opus achieves a 5.6 percentage-point gain for almost no change in cost, while Sonnet gains 9.4 percentage points and actually becomes cheaper than running it alone.",{"_uid":2932,"anchor":18,"content":2933,"eyebrow":18,"component":1515},"36377ddb-c960-4af0-b458-22326cf156dd",{"type":28,"content":2934},[2935,2958,2963,2968,2973,2978,2983],{"type":31,"attrs":2936,"content":2937},{"textAlign":17},[2938,2940,2944,2946,2950,2952,2956],{"text":2939,"type":36},"The two lines don’t move the same way. Opus goes from 78.0% to 83.6% while its per-task cost barely moves, $0.24 to $0.25 – ",{"text":2941,"type":36,"marks":2942},"a gain of 5.6 percentage points",[2943],{"type":183},{"text":2945,"type":36}," – for almost no change in cost. Sonnet goes from 72.1% to 81.5% – ",{"text":2947,"type":36,"marks":2948},"a 9.4 percentage-point jump –",[2949],{"type":183},{"text":2951,"type":36}," and gets ",{"text":2953,"type":36,"marks":2954},"cheaper",[2955],{"type":183},{"text":2957,"type":36},", $0.16 down to $0.15. The harness made the smaller model both more accurate and less expensive than running it alone, which is to say our system does not trade accuracy for cost and maintains token efficiency.",{"type":31,"attrs":2959,"content":2960},{"textAlign":17},[2961],{"text":2962,"type":36},"The most likely reason for this is that the planner spends a small, fixed number of tokens up front to turn the task into milestones. This initial investment buys back far more downstream. Because the plan is clear, the executor minimizes its total steps. It wanders less, gets stuck in fewer loops, and abandons dead-end approaches much sooner.",{"type":31,"attrs":2964,"content":2965},{"textAlign":17},[2966],{"text":2967,"type":36},"Seeing the weaker model gain so much ground is also revealing. A lot of what looks like a raw capability gap between two models is actually just vulnerability to minor failure modes. A smaller model is more likely to fail a task because it cannot recover from a tool error, tries to retype a document instead of reading the file, or refuses to stop when a UI approach fails. When the scaffolding solves these mechanical roadblocks, the underlying reasoning abilities of the two models converge. They are much closer in actual capability than their solo scores suggest.",{"type":31,"attrs":2969,"content":2970},{"textAlign":17},[2971],{"text":2972,"type":36},"Across our two runs, the full Sonnet system reaches about 98% of the Opus system’s score at roughly 43% of the total cost. The per-task figures above count only output tokens; this 43% is a fuller measure that includes all input and output compute across the run, which is why the gap between the two models is wider here than the output-only numbers suggest. For a benchmark, you just report the highest number. For production, this tradeoff is everything. Every task runs on a budget, and the question is never “what is most capable?” but rather “what is the cheapest model that can reliably complete this work?”",{"type":31,"attrs":2974,"content":2975},{"textAlign":17},[2976],{"text":2977,"type":36},"This is why we treat the model as a swappable component. The harness provides structure. The model is simply a setting you tune based on the complexity of the work and the constraints of your budget.",{"type":911,"attrs":2979,"content":2980},{"level":2580,"textAlign":17},[2981],{"text":2982,"type":36},"Knowing what it can’t do",{"type":31,"attrs":2984,"content":2985},{"textAlign":17},[2986],{"text":2987,"type":36},"The feasibility gate was evaluated on the benchmark’s 28 deliberately impossible tasks. It successfully identified 24 of them while incorrectly flagging only 2 of the 333 feasible tasks as impossible. This translates to an 85.7% recall on infeasible tasks and a 99.4% specificity on valid ones.",{"_uid":2989,"image":2990,"layout":1828,"caption":3001,"component":1830},"dfa46780-90fc-49f5-b33c-82e607da972b",[2991],{"_uid":2992,"image":2993,"video":2997,"lottie":2999,"component":203},"c71990bc-7801-4b5b-8140-ac329d17e81d",{"id":2994,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2995,"copyright":18,"fieldtype":19,"meta_data":2996,"is_external_url":198},180561820012647,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F3760x1556\u002F6700f42e5a\u002Fconfusion-matrix.png",{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":2998},{},{"id":17,"alt":17,"name":18,"focus":17,"title":17,"source":17,"filename":18,"copyright":17,"fieldtype":19,"meta_data":3000},{},"The feasibility gate evaluated across OSWorld’s task distribution. It accurately caught 24 impossible tasks while only falsely rejecting 2 valid ones, maintaining high specificity so real work is not abandoned.",{"_uid":3003,"anchor":18,"content":3004,"eyebrow":18,"component":1515},"b4c84777-476f-4741-9736-9528c0b0383c",{"type":28,"content":3005},[3006,3011,3022,3027,3032,3043,3048,3053],{"type":31,"attrs":3007,"content":3008},{"textAlign":17},[3009],{"text":3010,"type":36},"Tuning this threshold is difficult. It is easy to build a highly cautious agent that catches every impossible request, but it will inevitably abandon valid work. On the other hand, an agent optimized purely for task completion will fabricate results when it hits a dead end. The hard part is catching the genuine dead-ends without crying wolf on the real ones.",{"type":31,"attrs":3012,"content":3013},{"textAlign":17},[3014,3016,3020],{"text":3015,"type":36},"The four impossible tasks the gate failed to catch stem from a mismatch between real-world computing and benchmark constraints. The model attempted these tasks because, in a standard environment, ",{"text":3017,"type":36,"marks":3018},"they are possible",[3019],{"type":183},{"text":3021,"type":36},". A human user would solve them by simply installing an extension, downloading a plugin, or updating the software.",{"type":31,"attrs":3023,"content":3024},{"textAlign":17},[3025],{"text":3026,"type":36},"The agent’s pre-trained knowledge correctly recognizes these workarounds as valid. It only fails because OSWorld relies on a tightly restricted operating system where external downloads are blocked. The model’s reasoning was sound for a real computer; it failed only because the sandbox blocks the workarounds a production environment would allow.",{"type":911,"attrs":3028,"content":3029},{"level":2580,"textAlign":17},[3030],{"text":3031,"type":36},"Superscore",{"type":31,"attrs":3033,"content":3034},{"textAlign":17},[3035,3037,3041],{"text":3036,"type":36},"During internal testing we ran the system many times over. If we take the best result on each task across all our runs with Opus 4.7, ",{"text":3038,"type":36,"marks":3039},"the agent scores 90.9%",[3040],{"type":183},{"text":3042,"type":36}," – 317 of the 361 tasks solved completely, and another 13 solved partially. We call this the superscore. With Sonnet it is 87.6%, over a larger number of runs (8 for Opus, 14 for Sonnet).",{"type":31,"attrs":3044,"content":3045},{"textAlign":17},[3046],{"text":3047,"type":36},"Unlike our headline 83.6% from a single run, this composite score is not something we would claim on a leaderboard. However, it does serve as a diagnostic tool, proving that the underlying model possesses the reasoning, tool priors, and multi-modal understanding required to solve 9 out of 10 tasks.",{"type":31,"attrs":3049,"content":3050},{"textAlign":17},[3051],{"text":3052,"type":36},"The 7% gap between our single-run score and the superscore is entirely variance. When the agent fails a task that it proves capable of solving in a subsequent run, the failure is rarely a lack of intelligence. It is usually a mechanical execution error.",{"type":31,"attrs":3054,"content":3055},{"textAlign":17},[3056],{"text":3057,"type":36},"The point we want to highlight is that when a system can solve over 90% of an evaluation, the ceiling is in sight. The core problem is no longer figuring out if a model is smart enough to do the work. It shifts to building guardrails, self-correction loops, and independent verification systems that guarantee the model does the work correctly on the very first attempt.",{"_uid":3059,"anchor":3060,"content":3061,"eyebrow":18,"component":1515},"82b9616a-6ad5-4df1-a576-74b6d83a2455","Interesting findings",{"type":28,"content":3062},[3063,3067,3072,3077,3082,3087,3092,3097,3102,3113,3152,3157,3162,3167,3178,3183,3188,3193,3198,3203,3208,3213,3218,3241,3246,3251],{"type":911,"attrs":3064,"content":3065},{"level":1468,"textAlign":17},[3066],{"text":3060,"type":36},{"type":31,"attrs":3068,"content":3069},{"textAlign":17},[3070],{"text":3071,"type":36},"A lot of what we learned came from watching the agent run thousands of times. We noticed finer details in its behavior that we didn’t expect. We turned some into architectural improvements, while others revealed interesting quirks about how these models behave in the wild.",{"type":911,"attrs":3073,"content":3074},{"level":2580,"textAlign":17},[3075],{"text":3076,"type":36},"Sonnet’s “just do it” problem",{"type":31,"attrs":3078,"content":3079},{"textAlign":17},[3080],{"text":3081,"type":36},"Sonnet does not like to give up. On a task to extract hidden audio from an image – a task that was deliberately impossible, because there was no hidden audio – it worked through every steganography tool it could find, decided the file must be password-protected, and then installed a password cracker and a wordlist to brute-force a password that didn’t exist.",{"type":31,"attrs":3083,"content":3084},{"textAlign":17},[3085],{"text":3086,"type":36},"On another task, a website blocked it (our tasks run on AWS, which web servers often treat as a scraper), and its solution was to install Tor to route around the block. We found out about that one when AWS emailed us at 1 A.M.",{"type":31,"attrs":3088,"content":3089},{"textAlign":17},[3090],{"text":3091,"type":36},"We started calling this the “just do it” mentality: the model will keep going long after a person would have stopped and asked whether the task even makes sense. In the real world it is the most dangerous trait by far. An agent operating real systems needs to know when to stop and escalate the problem back to a person, and the instinct to find a way no matter what is exactly the instinct you don’t want.",{"type":31,"attrs":3093,"content":3094},{"textAlign":17},[3095],{"text":3096,"type":36},"This is the other side of the feasibility gate from earlier: knowing when not to start is one half, and knowing when to stop is the other. We have a lot more to say about this in a subsequent post.",{"type":911,"attrs":3098,"content":3099},{"level":2580,"textAlign":17},[3100],{"text":3101,"type":36},"Phantom tools",{"type":31,"attrs":3103,"content":3104},{"textAlign":17},[3105,3107,3111],{"text":3106,"type":36},"Across both runs, about 0.5% of the agent’s tool calls – 57 out of 11,603 – were calls to tools we never gave it. These were not random, though. They were standard computer use primitives, things like ",{"text":3108,"type":36,"marks":3109},"triple_click",[3110],{"type":2617},{"text":3112,"type":36},", ingrained in model weights via heavy reinforcement during computer use training. This prior is so strong that the model reaches for these tools even when they are absent from the provided schema.",{"type":31,"attrs":3114,"content":3115},{"textAlign":17},[3116,3118,3121,3123,3127,3129,3133,3135,3139,3140,3144,3146,3150],{"text":3117,"type":36},"Interestingly, when we returned an error saying ",{"text":3108,"type":36,"marks":3119},[3120],{"type":2617},{"text":3122,"type":36}," doesn’t exist, the model tried again with ",{"text":3124,"type":36,"marks":3125},"triple_click_safe",[3126],{"type":2617},{"text":3128,"type":36},", then ",{"text":3130,"type":36,"marks":3131},"triple_click_replacement",[3132],{"type":2617},{"text":3134,"type":36},", ",{"text":3136,"type":36,"marks":3137},"triple_click_workaround",[3138],{"type":2617},{"text":3134,"type":36},{"text":3141,"type":36,"marks":3142},"triple_click_substitute",[3143],{"type":2617},{"text":3145,"type":36},", and at one point ",{"text":3147,"type":36,"marks":3148},"triple_click_does_not_exist",[3149],{"type":2617},{"text":3151,"type":36},". It was confident enough the capability should exist that it kept permuting the name looking for one we’d accept, rather than concluding the tool wasn’t there.",{"type":31,"attrs":3153,"content":3154},{"textAlign":17},[3155],{"text":3156,"type":36},"We read this as a small window into a tension in heavily RL’d models. The training instills tool-use habits strong enough to override the actual tools provided in context.",{"type":31,"attrs":3158,"content":3159},{"textAlign":17},[3160],{"text":3161,"type":36},"It didn’t cost us anything measurable, but it’s a reminder that the model arrives with priors about the tools present in the environment it was trained in that don’t always match the tools in the environment it is operating in.",{"type":911,"attrs":3163,"content":3164},{"level":2580,"textAlign":17},[3165],{"text":3166,"type":36},"Letting a model read a file instead of writing code",{"type":31,"attrs":3168,"content":3169},{"textAlign":17},[3170,3172,3176],{"text":3171,"type":36},"Early on we noticed the agent doing something wasteful. To look at the contents of a file, it would open a terminal, write a few lines of Python, run them, and read the output. It clearly ",{"text":3173,"type":36,"marks":3174},"wanted",[3175],{"type":1481},{"text":3177,"type":36}," the file’s contents in front of it, and was writing code as the means to get there. The text contents of the file constitute far fewer tokens than an image containing the same information.",{"type":31,"attrs":3179,"content":3180},{"textAlign":17},[3181],{"text":3182,"type":36},"We gave it a more direct route: tools that read a file and hand the contents straight back, with PDFs and images passed in natively as input the model can see. The behavior was telling us what tool it wished it had. Once we built it, the detour disappeared. We think this is a general pattern worth paying attention to – when a model keeps doing something roundabout, it’s often describing a missing tool, and the fix is to listen rather than to prompt it out of the habit.",{"type":911,"attrs":3184,"content":3185},{"level":2580,"textAlign":17},[3186],{"text":3187,"type":36},"Two strikes, then switch",{"type":31,"attrs":3189,"content":3190},{"textAlign":17},[3191],{"text":3192,"type":36},"A model will try the same thing more times than a person would. The agent would frequently fail at an approach, try it again with a slight variation, and fail again, burning through its step budget on a dead end.",{"type":31,"attrs":3194,"content":3195},{"textAlign":17},[3196],{"text":3197,"type":36},"So, we gave the executor a rule: two failed attempts at the same mechanism is the signal to switch approaches entirely. This small addition to the system prompt changed the shape of our trajectories.",{"type":31,"attrs":3199,"content":3200},{"textAlign":17},[3201],{"text":3202,"type":36},"The interesting thing to us is that this had to be said at all. The instinct to keep trying the same door is strong in these models, and a surprising amount of getting good performance is teaching the agent when to stop doing the thing it wants to keep doing. It’s the same lesson as the feasibility gate, one level down.",{"type":911,"attrs":3204,"content":3205},{"level":2580,"textAlign":17},[3206],{"text":3207,"type":36},"Where the evaluator falls short",{"type":31,"attrs":3209,"content":3210},{"textAlign":17},[3211],{"text":3212,"type":36},"The most useful thing we learned from this benchmark is that grading the work is itself a verification problem – and OSWorld’s grader, built with every advantage, was still wrong often enough to matter.",{"type":31,"attrs":3214,"content":3215},{"textAlign":17},[3216],{"text":3217,"type":36},"In a benchmark, the task is fixed and known in advance, the correct end state is decided up front, and the grader is written by people who can see exactly what they’re checking for. Even with all of that, we kept hitting cases the grader was wrong:",{"type":1580,"content":3219},[3220,3227,3234],{"type":1583,"content":3221},[3222],{"type":31,"attrs":3223,"content":3224},{"textAlign":17},[3225],{"text":3226,"type":36},"Golden files that were themselves wrong. Roughly 10% of the reference outputs we examined had errors (e.g., a misspelled column name, a column sorted the wrong way) making the task impossible to pass, because the answer it was graded against was incorrect.",{"type":1583,"content":3228},[3229],{"type":31,"attrs":3230,"content":3231},{"textAlign":17},[3232],{"text":3233,"type":36},"Valid work it couldn’t recognize. The agent reached the right end state by an unorthodox route, or through a value the grader expected to be hard-coded, and the check failed because it was looking for one specific path.",{"type":1583,"content":3235},[3236],{"type":31,"attrs":3237,"content":3238},{"textAlign":17},[3239],{"text":3240,"type":36},"Answers that had gone stale. Some tasks depend on live data. The agent retrieves the current value, the golden file holds an old one, and the two no longer match.",{"type":31,"attrs":3242,"content":3243},{"textAlign":17},[3244],{"text":3245,"type":36},"We built our own verifier to see how much of this was the grader rather than the agent. It’s an agent that checks the executor’s work against the actual end state of the machine instead of against a stored answer, and in our experiments it was often the more reliable judge of whether a task was truly done. But the rigidity of the evaluator capped how much it could move our reported score. When the grader itself is the source of truth, a better judge can’t get credit for being right. We think a flexible verifier is a more interesting system than this benchmark can show, and we’ll cover it in a later post.",{"type":31,"attrs":3247,"content":3248},{"textAlign":17},[3249],{"text":3250,"type":36},"The reason we keep returning to this is what the evaluator’s errors actually imply. A benchmark like OSWorld is the friendliest possible setting for verification: the task is fixed, the correct end state is settled up front, and the grader was written by people who could see exactly what they were checking.",{"type":31,"attrs":3252,"content":3253},{"textAlign":17},[3254],{"text":3255,"type":36},"Even here, the grader was wrong often enough to matter. The bugs themselves are easy to fix; the lesson is harder to dismiss. If verification is this difficult when the answer is known in advance and someone wrote the check by hand, it does not get easier anywhere else.",{"_uid":3257,"anchor":3258,"content":3259,"eyebrow":18,"component":1515},"a5237648-7c75-4dad-abc8-61f966b344b3","Beyond the benchmark",{"type":28,"content":3260},[3261,3265,3270,3275,3280,3285],{"type":911,"attrs":3262,"content":3263},{"level":1468,"textAlign":17},[3264],{"text":3258,"type":36},{"type":31,"attrs":3266,"content":3267},{"textAlign":17},[3268],{"text":3269,"type":36},"We had a lot of fun with this, and we learned an enormous amount watching the agent run thousands of times – about the models, and the small things around them that turn out to matter more than raw capability.",{"type":31,"attrs":3271,"content":3272},{"textAlign":17},[3273],{"text":3274,"type":36},"This benchmark is of course just one signal. It tells you the agent can do the work under clear-cut, pre-defined conditions. But it does not tell you whether the same agent holds up against work that is open-ended and consequential.",{"type":31,"attrs":3276,"content":3277},{"textAlign":17},[3278],{"text":3279,"type":36},"That second question is where we spend our time. The work we do in production lives in finance and operations, where the tasks are longer, systems are messier, and the actual information you need is scattered across places that were never meant to talk to each other. A system that succeeds here has to assemble that scattered context into an accurate picture before it acts, operate under real constraints on what it is allowed to do, and produce evidence that its work is correct rather than merely plausible. You have to be able to stand behind the work without an answer key in front of you.",{"type":31,"attrs":3281,"content":3282},{"textAlign":17},[3283],{"text":3284,"type":36},"This is why we treat a state-of-the-art score as the baseline. A top result proves the underlying models are finally capable enough to do this work. The scaffolding that makes capability reliable enough to put into production, and the verification that proves it worked, is the harder and more interesting problem we spend more time on. It’s also why we bother topping a leaderboard at all: the people we build for should have the best system in the world operating their work, and the benchmark keeps us honest about that.",{"type":31,"attrs":3286,"content":3287},{"textAlign":17},[3288,3290,3295],{"text":3289,"type":36},"If any of that sounds like the thing you want to work on, ",{"text":3291,"type":36,"marks":3292},"we’re hiring",[3293],{"type":256,"attrs":3294},{"href":2192,"uuid":17,"anchor":17,"target":17,"linktype":172},{"text":3296,"type":36},". We’re especially interested in people thinking about computer use, verifying agent work, self-improving systems, and coherence over long horizons.",{"_uid":3298,"anchor":3299,"content":3300,"eyebrow":18,"component":1515},"a3086b58-329e-4387-986c-14162efdaa45","Appendix",{"type":28,"content":3301},[3302,3306,3311],{"type":911,"attrs":3303,"content":3304},{"level":1468,"textAlign":17},[3305],{"text":3299,"type":36},{"type":31,"attrs":3307,"content":3308},{"textAlign":17},[3309],{"text":3310,"type":36},"Here are some resources and helpful links:",{"type":1580,"content":3312},[3313,3322,3333,3397,3408],{"type":1583,"content":3314},[3315],{"type":31,"attrs":3316,"content":3317},{"textAlign":17},[3318],{"text":2296,"type":36,"marks":3319},[3320],{"type":256,"attrs":3321},{"href":2300,"uuid":17,"anchor":17,"target":17,"linktype":172},{"type":1583,"content":3323},[3324],{"type":31,"attrs":3325,"content":3326},{"textAlign":17},[3327],{"text":3328,"type":36,"marks":3329},"Our harness",[3330],{"type":256,"attrs":3331},{"href":3332,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fgithub.com\u002FPointer-so\u002FOSWorld",{"type":1583,"content":3334},[3335,3340],{"type":31,"attrs":3336,"content":3337},{"textAlign":17},[3338],{"text":3339,"type":36},"Trajectories:",{"type":1580,"content":3341},[3342,3353,3364,3375,3386],{"type":1583,"content":3343},[3344],{"type":31,"attrs":3345,"content":3346},{"textAlign":17},[3347],{"text":3348,"type":36,"marks":3349},"Verified Opus 4.7 – 83.6%",[3350],{"type":256,"attrs":3351},{"href":3352,"uuid":17,"anchor":17,"target":17,"linktype":172},"http:\u002F\u002Fcdn.pointer.ai\u002Fartifacts\u002Fopus-4-7-traj_verified.zip",{"type":1583,"content":3354},[3355],{"type":31,"attrs":3356,"content":3357},{"textAlign":17},[3358],{"text":3359,"type":36,"marks":3360},"Verified Sonnet 4.6 – 81.5%",[3361],{"type":256,"attrs":3362},{"href":3363,"uuid":17,"anchor":17,"target":17,"linktype":172},"http:\u002F\u002Fcdn.pointer.ai\u002Fartifacts\u002Fsonnet-4-6-traj_verified.zip",{"type":1583,"content":3365},[3366],{"type":31,"attrs":3367,"content":3368},{"textAlign":17},[3369],{"text":3370,"type":36,"marks":3371},"Internal Opus 4.7 – 84.7%",[3372],{"type":256,"attrs":3373},{"href":3374,"uuid":17,"anchor":17,"target":17,"linktype":172},"http:\u002F\u002Fcdn.pointer.ai\u002Fartifacts\u002Fopus-4-7-traj_0.zip",{"type":1583,"content":3376},[3377],{"type":31,"attrs":3378,"content":3379},{"textAlign":17},[3380],{"text":3381,"type":36,"marks":3382},"Internal Opus 4.7 – 82.0%",[3383],{"type":256,"attrs":3384},{"href":3385,"uuid":17,"anchor":17,"target":17,"linktype":172},"http:\u002F\u002Fcdn.pointer.ai\u002Fartifacts\u002Fopus-4-7-traj_1.zip",{"type":1583,"content":3387},[3388],{"type":31,"attrs":3389,"content":3390},{"textAlign":17},[3391],{"text":3392,"type":36,"marks":3393},"Internal Opus 4.7 – 84.1%",[3394],{"type":256,"attrs":3395},{"href":3396,"uuid":17,"anchor":17,"target":17,"linktype":172},"http:\u002F\u002Fcdn.pointer.ai\u002Fartifacts\u002Fopus-4-7-traj_2.zip",{"type":1583,"content":3398},[3399],{"type":31,"attrs":3400,"content":3401},{"textAlign":17},[3402],{"text":3403,"type":36,"marks":3404},"Claude Opus 4.7 system card",[3405],{"type":256,"attrs":3406},{"href":3407,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fcdn.sanity.io\u002Ffiles\u002F4zrzovbb\u002Fwebsite\u002F037f06850df7fbe871e206dad004c3db5fd50340.pdf",{"type":1583,"content":3409},[3410],{"type":31,"attrs":3411,"content":3412},{"textAlign":17},[3413],{"text":3414,"type":36,"marks":3415},"Claude Sonnet 4.6 system card",[3416],{"type":256,"attrs":3417},{"href":3418,"uuid":17,"anchor":17,"target":17,"linktype":172},"https:\u002F\u002Fwww-cdn.anthropic.com\u002Fbbd8ef16d70b7a1665f14f306ee88b53f686aa75.pdf",{"id":2268,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":2269,"copyright":18,"fieldtype":19,"meta_data":3420,"is_external_url":198},{},{"type":28,"content":3422},[3423],{"type":31,"attrs":3424,"content":3425},{"textAlign":17},[3426],{"text":106,"type":36,"marks":3427},[3428],{"type":183},{"name":3430,"created_at":3431,"published_at":3432,"updated_at":3433,"id":3434,"uuid":3435,"content":3436,"slug":3441,"full_slug":3442,"sort_by_date":17,"position":1438,"tag_list":3443,"is_startpage":198,"parent_id":2228,"meta_data":17,"group_id":3444,"first_published_at":3432,"release_id":17,"lang":418,"path":17,"alternates":3445,"default_full_slug":17,"translated_slugs":17,"_stopResolving":109},"Pulkit Arya","2026-05-21T21:36:52.627Z","2026-06-25T19:50:19.199Z","2026-06-25T19:50:19.215Z",179035392554938,"461e9f60-a349-48bd-a0f0-27c9569ca034",{"Name":3430,"_uid":2218,"Headshot":3437,"component":2224},{"id":3438,"alt":18,"name":18,"focus":18,"title":18,"source":18,"filename":3439,"copyright":18,"fieldtype":19,"meta_data":3440,"is_external_url":198},179037598312364,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F286886172940351\u002F1187x1187\u002Fee8f3a562f\u002Fpulkit.jpg",{},"pulkit-arya","blog\u002Fauthors\u002Fpulkit-arya",[],"bbf91eb1-f879-4c67-80c8-279b1a1aea9c",[],{"type":28,"content":3447},[3448],{"type":31,"attrs":3449,"content":3450},{"textAlign":17},[3451],{"text":3452,"type":36},"We just hit a new state of the art on OSWorld. Here's how we built the agent, what surprised us along the way, and why topping the benchmark is just the starting line.","2026-05-26 00:00",[],-90,[],"e3cd778a-38d3-4d5a-985e-3a1d7a0671e9","2026-06-25T19:48:20.060Z",[],1789673308756]