[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-what-a-coding-agent-forgets-when-it-plays-arc-agi-3":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},10873,"what-a-coding-agent-forgets-when-it-plays-arc-agi-3","What a Coding Agent Forgets When It Plays ARC-AGI-3","Researchers tracked a coding agent's own files across ARC-AGI-3 and found it ditches old code but never edits its notes, letting contradictions pile up.","A coding agent playing an abstract puzzle benchmark rewrites its own playbook almost every level - and never bothers to clean up the notes it leaves behind.\n\nResearchers wired a frozen language model into a harness that lets it write and run Python and shell scripts to play ARC-AGI-3, a set of interactive reasoning games that come with no instructions. The agent carries no memory between turns except what it writes to disk, so every script and note becomes a record of what it currently believes. Across seven runs on three model backbones from two model families, the team traced each of those beliefs from the moment it formed to the moment it was corrected or dropped. Out of 630 references to earlier scripts, only 33 crossed into a new level, and the agent abandoned 74% of its code once a level ended, rewriting general rules into fresh scripts while dropping level-specific details.\n\nThat is a cleaner kind of forgetting than the field usually worries about. Catastrophic forgetting in neural networks wipes out old knowledge indiscriminately; here the agent's complete action log lets it forget selectively, while its text notes just accumulate claims without ever being edited, leaving contradictions to pile up until the log settles them. The costliest mistake was not losing knowledge - it was holding onto a hard-coded value past the point where it stopped being true.\n\nIt is a reminder that for agents built on frozen models, memory management is a file-system problem long before it is a machine-learning one.","[\"ai agents\",\"arc-agi\",\"continual learning\",\"llm research\"]","2026-10-09T04:00:00.000Z","2026-10-09T19:26:00.671Z","2026-10-09T19:26:05.152Z","published",null,[],"ai",[26,27,28,29],"ai agents","arc-agi","continual learning","llm research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.11450",0,{"sections":36},[37,40,44,49,54,59,63,68,73,78,83,88,93,98],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",6619,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",926,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",486,"2026-10-08T22:40:11.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",474,"2026-10-08T22:00:00.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",229,"2026-10-08T20:47:10.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",192,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",181,"2026-10-08T23:26:35.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Startups","startups",117,"2026-10-08T16:45:00.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Software","software",114,"2026-10-08T17:57:01.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Dev Tools","dev-tools",105,"2026-10-07T16:59:11.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"General","general",66,"2026-10-09T04:46:11.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"Gaming","gaming",58,"2026-10-08T20:08:45.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",34,"2026-10-08T14:00:22.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]