[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-video-rsi-lets-ai-agents-rewrite-their-own-analysis-tools":10,"sections":45},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":35,"tags":36,"sources":40,"feedback":44,"feedback_at":22,"cost_usd":44,"total_tokens":44},8535,"video-rsi-lets-ai-agents-rewrite-their-own-analysis-tools","Video-RSI Lets AI Agents Rewrite Their Own Analysis Tools","A new framework lets AI video agents diagnose their own failures by rewatching training footage, then rewrite the tools they use to watch video.","A new AI framework lets video-understanding agents catch their own blind spots and rewrite the code that controls what they watch.\n\nThe framework, called Video-RSI, comes with a public code release on GitHub. Video agents work through a harness, the executable layer that decides which frames they look at and how they interpret them. Normally an agent stuck with a failure trace has no way to test other explanations for why it got something wrong. Video-RSI closes that gap by sending the agent's own language model back to the original training videos, where it gathers new observations and weighs competing explanations before rewriting its harness. A cost-aware selection step then decides which revisions to keep, weighing answer accuracy against how many frames the change requires the agent to process.\n\nThat's a meaningful shift from typical self-improving agents, which mostly refine themselves using only the evidence a fixed pipeline happened to record. Letting an agent go back and investigate, rather than just reread its own homework, is a plausible route to agents that improve without needing a human to relabel their mistakes. The tradeoff the paper is chasing, better accuracy using fewer frames, matters because frame-by-frame video processing is exactly the kind of compute cost that makes video AI expensive to run at scale.\n\nIt's still an academic result tested against the researchers' own benchmarks, not a product running in the wild, so the real test is whether harness evolution holds up outside curated evaluation settings.","[\"ai\",\"video-understanding\",\"ai-agents\",\"self-improvement\"]","2026-09-30T04:00:00.000Z","2026-09-30T09:42:39.614Z","2026-09-30T09:42:43.307Z","published",null,[24,30],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"State what the cited benchmark results actually show (or note the paper provides no concrete accuracy\u002Fframe-count figures) instead of asserting vague 'got more answers right' claims, and replace the trailing caveat-only closing paragraph with an ending that adds context rather than just trailing off on caveats.","resolved",{"id":31,"reviewer":32,"round":33,"reason":34,"status":29},"publisher-r2","publisher",2,"The dek and closing body paragraph openly flag that the article lacks the concrete accuracy\u002Fframe-count numbers behind its central claims, so key facts are missing rather than merely unstated background.","ai",[35,37,38,39],"video-understanding","ai-agents","self-improvement",[41],{"name":42,"url":43},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.37950",0,{"sections":46},[47,50,54,58,63,68,73,78,83,87,92,97,102,107],{"name":48,"slug":35,"count":49,"latest_published_at":18},"AI",5105,{"name":51,"slug":52,"count":53,"latest_published_at":18},"Security","security",785,{"name":55,"slug":56,"count":57,"latest_published_at":18},"Policy","policy",417,{"name":59,"slug":60,"count":61,"latest_published_at":62},"Deals","deals",284,"2026-09-29T21:00:00.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Hardware","hardware",194,"2026-09-29T13:16:04.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Consumer Tech","consumer-tech",142,"2026-09-29T18:38:03.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":18},"Dev Tools","dev-tools",90,{"name":88,"slug":89,"count":90,"latest_published_at":91},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":103,"slug":104,"count":105,"latest_published_at":106},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":108,"slug":109,"count":110,"latest_published_at":111},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]