Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456 Paper page - Learning to Configure Agentic AI Systems
Please give a thumbs up to this comment if you found it helpful!
\n
If you want recommendations for any Paper on Hugging Face checkout this Space
\n
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend
\n","updatedAt":"2026-02-18T01:39:06.783Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7128006815910339},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2602.11574","authors":[{"_id":"698f24193ae80e6a12af8e20","user":{"_id":"657a33bb06e44e4565422dfa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/657a33bb06e44e4565422dfa/WA1wznlOFKpmmsePIDnzq.jpeg","isPro":false,"fullname":"Aditya Taparia","user":"aditya-taparia","type":"user"},"name":"Aditya Taparia","status":"claimed_verified","statusLastChangedAt":"2026-02-17T15:51:35.020Z","hidden":false},{"_id":"698f24193ae80e6a12af8e21","user":{"_id":"6584c10444b9961f765a776d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6584c10444b9961f765a776d/KDvYnmKk6mQFJha3otq17.png","isPro":false,"fullname":"Som Sagar","user":"sssagar","type":"user"},"name":"Som Sagar","status":"claimed_verified","statusLastChangedAt":"2026-02-18T09:06:50.937Z","hidden":false},{"_id":"698f24193ae80e6a12af8e22","name":"Ransalu Senanayake","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/657a33bb06e44e4565422dfa/xKbDrhSoOKsD1Koy4-Dvo.jpeg"],"publishedAt":"2026-02-12T04:45:44.000Z","submittedOnDailyAt":"2026-02-17T17:00:55.832Z","title":"Learning to Configure Agentic AI Systems","submittedOnDailyBy":{"_id":"657a33bb06e44e4565422dfa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/657a33bb06e44e4565422dfa/WA1wznlOFKpmmsePIDnzq.jpeg","isPro":false,"fullname":"Aditya Taparia","user":"aditya-taparia","type":"user"},"summary":"Configuring LLM-based agent systems involves choosing workflows, tools, token budgets, and prompts from a large combinatorial design space, and is typically handled today by fixed large templates or hand-tuned heuristics. This leads to brittle behavior and unnecessary compute, since the same cumbersome configuration is often applied to both easy and hard input queries. We formulate agent configuration as a query-wise decision problem and introduce ARC (Agentic Resource & Configuration learner), which learns a light-weight hierarchical policy using reinforcement learning to dynamically tailor these configurations. Across multiple benchmarks spanning reasoning and tool-augmented question answering, the learned policy consistently outperforms strong hand-designed and other baselines, achieving up to 25% higher task accuracy while also reducing token and runtime costs. These results demonstrate that learning per-query agent configurations is a powerful alternative to \"one size fits all\" designs.","upvotes":14,"discussionId":"698f241a3ae80e6a12af8e23","githubRepo":"https://github.com/somsagar07/Context_Optimization","githubRepoAddedBy":"user","ai_summary":"Learning per-query agent configurations through reinforcement learning improves task accuracy while reducing computational costs compared to fixed templates and hand-tuned heuristics.","ai_keywords":["LLM-based agent systems","reinforcement learning","hierarchical policy","query-wise decision problem","agent configuration","token budget","prompt engineering","tool-augmented question answering","reasoning tasks","task accuracy","computational efficiency"],"githubStars":6,"organization":{"_id":"6994c024db3cbf241bd24b0b","name":"lens-lab-AI","fullname":"LENS Lab","avatar":"https://cdn-uploads.huggingface.co/production/uploads/657a33bb06e44e4565422dfa/hOk6Tv7V7OSECOvyk_lOU.webp"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"657a33bb06e44e4565422dfa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/657a33bb06e44e4565422dfa/WA1wznlOFKpmmsePIDnzq.jpeg","isPro":false,"fullname":"Aditya Taparia","user":"aditya-taparia","type":"user"},{"_id":"6584c10444b9961f765a776d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6584c10444b9961f765a776d/KDvYnmKk6mQFJha3otq17.png","isPro":false,"fullname":"Som Sagar","user":"sssagar","type":"user"},{"_id":"63fcdb5cd9818304e85068ce","avatarUrl":"/avatars/a37ffae72e121d7d81489a8bc84b0132.svg","isPro":false,"fullname":"V","user":"Sreevishakh","type":"user"},{"_id":"67db71df2aeb1103caf0eabc","avatarUrl":"/avatars/0eeff116d5289a67712ec738fb0f4424.svg","isPro":false,"fullname":"Athira Raghumadhavan","user":"athirarmadhavan","type":"user"},{"_id":"6994ca05cc4846580d88d691","avatarUrl":"/avatars/6695bfcff73a30f2cf3bcd7f7ab2f93b.svg","isPro":false,"fullname":"Meenakshi Rajesh","user":"meenakshirajesh1999","type":"user"},{"_id":"6994dbbd4cffa74430c768d9","avatarUrl":"/avatars/929f7f4ccdea05a693996d5daeb8bc44.svg","isPro":false,"fullname":"Akshay Jayasoorya","user":"ajsoorya","type":"user"},{"_id":"6808642584cac4b136e942a8","avatarUrl":"/avatars/e2a3c2c936d2e814d0e46529ff99c2b5.svg","isPro":false,"fullname":"Benhar John","user":"benharjohn","type":"user"},{"_id":"67967aa405c4a94ebd666a1a","avatarUrl":"/avatars/2c8fb537859428b69028234748641d36.svg","isPro":false,"fullname":"Nevin","user":"nevinselby","type":"user"},{"_id":"68b327e672075acbc766d04e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/18c9-o4npNoPUaRIeLEZZ.png","isPro":false,"fullname":"Son Nguyen","user":"snguye88","type":"user"},{"_id":"63f1b25dbc705ef8c23fc86e","avatarUrl":"/avatars/7d1036731e5334ba93f649e02547c959.svg","isPro":false,"fullname":"Emma","user":"liuuu121","type":"user"},{"_id":"6995341c942f907479c40297","avatarUrl":"/avatars/956129eb76c1ac847c01e95155836b2b.svg","isPro":false,"fullname":"Riana Chatterjee","user":"RianaChatterjee1","type":"user"},{"_id":"626299ddf66aed28cef2e2c6","avatarUrl":"/avatars/909ceefb1ca6725014b5d6c6977879e2.svg","isPro":false,"fullname":"Eren Sadikoglu","user":"ErenSadikoglu","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0,"organization":{"_id":"6994c024db3cbf241bd24b0b","name":"lens-lab-AI","fullname":"LENS Lab","avatar":"https://cdn-uploads.huggingface.co/production/uploads/657a33bb06e44e4565422dfa/hOk6Tv7V7OSECOvyk_lOU.webp"}}">
Learning per-query agent configurations through reinforcement learning improves task accuracy while reducing computational costs compared to fixed templates and hand-tuned heuristics.
AI-generated summary
Configuring LLM-based agent systems involves choosing workflows, tools, token budgets, and prompts from a large combinatorial design space, and is typically handled today by fixed large templates or hand-tuned heuristics. This leads to brittle behavior and unnecessary compute, since the same cumbersome configuration is often applied to both easy and hard input queries. We formulate agent configuration as a query-wise decision problem and introduce ARC (Agentic Resource & Configuration learner), which learns a light-weight hierarchical policy using reinforcement learning to dynamically tailor these configurations. Across multiple benchmarks spanning reasoning and tool-augmented question answering, the learned policy consistently outperforms strong hand-designed and other baselines, achieving up to 25% higher task accuracy while also reducing token and runtime costs. These results demonstrate that learning per-query agent configurations is a powerful alternative to "one size fits all" designs.
Building agentic systems is hard, but configuring them is even harder.
We all know the struggle: Which LLM should handle the planning? Which tool does it need? How much context is too much? What is the most effective workflow?
In our new paper, Learning to Configure Agentic AI Systems, we propose a framework (called ARC) that automates these decisions. Instead of manual trial-and-error, we use a Hierarchical Reinforcement Learning (HRL) algorithm to dynamically find the best configuration for a given input.