[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-new-trick-for-teaching-ai-agent-networks-to-learn-on-the-job":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},1652,"a-new-trick-for-teaching-ai-agent-networks-to-learn-on-the-job","A New Trick for Teaching AI Agent Networks to Learn on the Job","Skill-MAS claims to solve a long-standing tradeoff in multi-agent AI: letting powerful models accumulate experience without retraining them.","Multi-agent AI systems can now learn from past runs without touching the model weights.\n\nResearchers introduced Skill-MAS, a framework designed to sidestep a persistent problem in automatic multi-agent system design. Current approaches split into two camps: use a large, capable frozen model that repeats the same mistakes run after run, or train a smaller model to internalize lessons but sacrifice raw capability. Skill-MAS proposes a third option — store orchestration knowledge as a separate, updatable \"Meta-Skill\" that improves through a feedback loop without gradient updates. The system samples multiple behavioral paths per task, then selectively reflects on high-priority failures to extract reusable strategic principles.\n\nThe distinction matters because most real-world deployments use frontier models that are too large and expensive to retrain on every workflow. If an orchestration layer can accumulate and transfer lessons across tasks and even across different base models, teams get compounding returns without compounding GPU bills. The paper reports gains across four benchmarks and four distinct base LLMs, which at least suggests the approach is not tuned to a single model's quirks.\n\nThat said, \"extensive experiments\" on benchmarks is a long way from production. The AI research pipeline is littered with techniques that transfer well inside a paper and poorly outside it — and a method that claims strong transferability across unseen tasks is exactly the kind of claim worth watching closely when someone tries to ship it.","[\"ai\",\"multi-agent\",\"llm\",\"research\"]","2026-06-18T04:00:00.000Z","2026-06-19T09:11:17.710Z","2026-06-19T14:21:36.108Z","published",null,[],"ai",[24,26,27,28],"multi-agent","llm","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.18837",0,{"sections":35},[36,40,44,49,54,59,64,68,72,76,81,86,91,96],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",490,"2026-06-19T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":39},"Security","security",132,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",88,"2026-06-16T09:26:09.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Consumer Tech","consumer-tech",78,"2026-06-16T17:58:24.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",62,"2026-06-18T15:24:16.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Deals","deals",58,"2026-06-19T14:43:50.000Z",{"name":65,"slug":66,"count":62,"latest_published_at":67},"Software","software","2026-06-16T20:00:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":39},"Dev Tools","dev-tools",50,{"name":73,"slug":74,"count":75,"latest_published_at":18},"Science","science",38,{"name":77,"slug":78,"count":79,"latest_published_at":80},"Gaming","gaming",31,"2026-06-16T15:25:13.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"General","general",26,"2026-06-13T18:35:15.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Startups","startups",23,"2026-06-16T15:00:00.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",19,"2026-06-14T08:00:00.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]