As model capabilities continue to improve, we argue that the bottleneck for autonomous scientific discovery is shifting from prescribing agent workflows to designing agent environments: the resources, constraints, and interfaces that shape agent behavior. We frame this as environment engineering: building environments that amplify productive behaviors, such as open-ended exploration, systematic artifact management, and inter-agent collaboration, while suppressing harmful behaviors, such as reward hacking and high-friction human oversight. We present EurekAgent, an environment-engineered agent system for metric-driven autonomous scientific discovery. EurekAgent engineers the environment along four dimensions: permissions engineering for bounded agent execution and isolated evaluation; artifact engineering for filesystem and Git-based collaboration; budget engineering for budget-aware exploration; and human-in-the-loop engineering for easy human supervision and intervention. EurekAgent sets new state-of-the-art results on multiple mathematics, kernel engineering, and machine learning tasks, including new state-of-the-art 26-circle packing results discovered with less than $11 in total API cost. We open-source our code and results, and call for environment engineering as a core research direction for developing reliable autonomous research agents.</p>\n","updatedAt":"2026-06-12T02:32:33.512Z","author":{"_id":"660bf98c3336a7e128a0e918","avatarUrl":"/avatars/3e3f2886bd4a730ec19b13aecc99279f.svg","fullname":"Amy Xin","name":"amyxx2001","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9107796549797058},"editors":["amyxx2001"],"editorAvatarUrls":["/avatars/3e3f2886bd4a730ec19b13aecc99279f.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.13662","authors":[{"_id":"6a2b63024957fcdd3aac05a4","user":{"_id":"660bf98c3336a7e128a0e918","avatarUrl":"/avatars/3e3f2886bd4a730ec19b13aecc99279f.svg","isPro":false,"fullname":"Amy Xin","user":"amyxx2001","type":"user","name":"amyxx2001"},"name":"Amy Xin","status":"claimed_verified","statusLastChangedAt":"2026-06-12T06:58:05.748Z","hidden":false},{"_id":"6a2b63024957fcdd3aac05a5","user":{"_id":"68f5cfb57b23d048745eb8a9","avatarUrl":"/avatars/8fc6894470f36700410a38b7faef7805.svg","isPro":false,"fullname":"Jiening Siow","user":"Little-d1d1","type":"user","name":"Little-d1d1"},"name":"Jiening Siow","status":"claimed_verified","statusLastChangedAt":"2026-06-12T06:58:03.665Z","hidden":false},{"_id":"6a2b63024957fcdd3aac05a6","name":"Junjie Wang","hidden":false},{"_id":"6a2b63024957fcdd3aac05a7","name":"Zijun Yao","hidden":false},{"_id":"6a2b63024957fcdd3aac05a8","name":"Fanjin Zhang","hidden":false},{"_id":"6a2b63024957fcdd3aac05a9","name":"Jian Song","hidden":false},{"_id":"6a2b63024957fcdd3aac05aa","name":"Lei Hou","hidden":false},{"_id":"6a2b63024957fcdd3aac05ab","name":"Juanzi Li","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/660bf98c3336a7e128a0e918/uGTnhnGIBVUmA06M5Xszd.mp4"],"publishedAt":"2026-06-11T00:00:00.000Z","submittedOnDailyAt":"2026-06-12T00:00:00.000Z","title":"EurekAgent: Agent Environment Engineering is All You Need For Autonomous Scientific Discovery","submittedOnDailyBy":{"_id":"660bf98c3336a7e128a0e918","avatarUrl":"/avatars/3e3f2886bd4a730ec19b13aecc99279f.svg","isPro":false,"fullname":"Amy Xin","user":"amyxx2001","type":"user","name":"amyxx2001"},"summary":"LLM-based agents have shown increasing potential in automating scientific discovery. Given an optimizable metric and an execution environment, they can propose, validate, and iterate scientific solutions, and have produced results that outperform human-designed approaches. As model capabilities continue to improve, we argue that the bottleneck for autonomous scientific discovery is shifting from prescribing agent workflows to designing agent environments: the resources, constraints, and interfaces that shape agent behavior. We frame this as environment engineering: building environments that amplify productive behaviors, such as open-ended exploration, systematic artifact management, and inter-agent collaboration, while suppressing harmful behaviors, such as reward hacking and high-friction human oversight. We present EurekAgent, an environment-engineered agent system for metric-driven autonomous scientific discovery. EurekAgent engineers the environment along four dimensions: permissions engineering for bounded agent execution and isolated evaluation; artifact engineering for filesystem and Git-based collaboration; budget engineering for budget-aware exploration; and human-in-the-loop engineering for easy human supervision and intervention. EurekAgent sets new state-of-the-art results on multiple mathematics, kernel engineering, and machine learning tasks, including new state-of-the-art 26-circle packing results discovered with less than $11 in total API cost. We open-source our code and results, and call for environment engineering as a core research direction for developing reliable autonomous research agents.","upvotes":15,"discussionId":"6a2b63034957fcdd3aac05ac","githubRepo":"https://github.com/THU-Team-Eureka/EurekAgent","githubRepoAddedBy":"user","ai_summary":"Environment engineering enhances autonomous scientific discovery by designing structured agent environments that optimize behaviors like exploration and collaboration while mitigating issues such as reward hacking and human oversight friction, as demonstrated by the EurekAgent system that achieves state-of-the-art results across multiple domains with low computational costs.","ai_keywords":["environment engineering","autonomous scientific discovery","agent environments","permissions engineering","artifact engineering","budget engineering","human-in-the-loop engineering","EurekAgent","metric-driven discovery","reward hacking"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":2,"organization":{"_id":"64db4fc57266618e854318f4","name":"THU-KEG","fullname":"Knowledge Engineer Group @ Tsinghua University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c4b46e549be47af1aafcd/5atqdE9AUWvYAHm9FNkG_.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"660bf98c3336a7e128a0e918","avatarUrl":"/avatars/3e3f2886bd4a730ec19b13aecc99279f.svg","isPro":false,"fullname":"Amy Xin","user":"amyxx2001","type":"user"},{"_id":"667f7c70c723a498798bf1ba","avatarUrl":"/avatars/610d371ca844e77c18ffa04b79aff4b6.svg","isPro":false,"fullname":"Gilson Siqueira","user":"barateza","type":"user"},{"_id":"68f5cfb57b23d048745eb8a9","avatarUrl":"/avatars/8fc6894470f36700410a38b7faef7805.svg","isPro":false,"fullname":"Jiening Siow","user":"Little-d1d1","type":"user"},{"_id":"6407e5294edf9f5c4fd32228","avatarUrl":"/avatars/8e2d55460e9fe9c426eb552baf4b2cb0.svg","isPro":false,"fullname":"Stoney Kang","user":"sikang99","type":"user"},{"_id":"648c4b46e549be47af1aafcd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c4b46e549be47af1aafcd/YgOHbmUM2EDM-lb7GdqXz.jpeg","isPro":false,"fullname":"Zijun","user":"TranSirius","type":"user"},{"_id":"6613a7d542da659656d85d28","avatarUrl":"/avatars/d052f49b3ae62708c5bcdb4bc34ffc5a.svg","isPro":false,"fullname":"Fatty","user":"FattyFatty","type":"user"},{"_id":"66cdd285c51a915bd5f2d017","avatarUrl":"/avatars/14e5794307e4672b1b51d26b31227e0f.svg","isPro":false,"fullname":"Jiajie Zhang","user":"NeoZ123","type":"user"},{"_id":"64265a0ba5ec4a5cbc532cb1","avatarUrl":"/avatars/9c7fe642d510a3cb880b44ea8e129651.svg","isPro":false,"fullname":"Wang Zhitong","user":"tommywang721","type":"user"},{"_id":"5fad02872498c65d4119eea7","avatarUrl":"/avatars/52241108461d510c70e81ef3eed8ed8a.svg","isPro":false,"fullname":"Jian Song","user":"sjj","type":"user"},{"_id":"63d88b7a3130cadcaf8d30c2","avatarUrl":"/avatars/09c060315db6be266c39b0f05ed24027.svg","isPro":false,"fullname":"LU","user":"jojoUla","type":"user"},{"_id":"6a2bb48606799459d832d807","avatarUrl":"/avatars/c441554495b184cc6de80a66f3cc31fb.svg","isPro":false,"fullname":"Lei Hou","user":"HLGreener","type":"user"},{"_id":"625a5446f1063e7085d5178a","avatarUrl":"/avatars/5e78186f13f74b14e01583e06ff6c4dc.svg","isPro":false,"fullname":"Hao Peng","user":"Wesleythu","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"64db4fc57266618e854318f4","name":"THU-KEG","fullname":"Knowledge Engineer Group @ Tsinghua University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c4b46e549be47af1aafcd/5atqdE9AUWvYAHm9FNkG_.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.13662.md","query":{}}">
EurekAgent: Agent Environment Engineering is All You Need For Autonomous Scientific Discovery
Abstract
Environment engineering enhances autonomous scientific discovery by designing structured agent environments that optimize behaviors like exploration and collaboration while mitigating issues such as reward hacking and human oversight friction, as demonstrated by the EurekAgent system that achieves state-of-the-art results across multiple domains with low computational costs.
LLM-based agents have shown increasing potential in automating scientific discovery. Given an optimizable metric and an execution environment, they can propose, validate, and iterate scientific solutions, and have produced results that outperform human-designed approaches. As model capabilities continue to improve, we argue that the bottleneck for autonomous scientific discovery is shifting from prescribing agent workflows to designing agent environments: the resources, constraints, and interfaces that shape agent behavior. We frame this as environment engineering: building environments that amplify productive behaviors, such as open-ended exploration, systematic artifact management, and inter-agent collaboration, while suppressing harmful behaviors, such as reward hacking and high-friction human oversight. We present EurekAgent, an environment-engineered agent system for metric-driven autonomous scientific discovery. EurekAgent engineers the environment along four dimensions: permissions engineering for bounded agent execution and isolated evaluation; artifact engineering for filesystem and Git-based collaboration; budget engineering for budget-aware exploration; and human-in-the-loop engineering for easy human supervision and intervention. EurekAgent sets new state-of-the-art results on multiple mathematics, kernel engineering, and machine learning tasks, including new state-of-the-art 26-circle packing results discovered with less than $11 in total API cost. We open-source our code and results, and call for environment engineering as a core research direction for developing reliable autonomous research agents.
Community
As model capabilities continue to improve, we argue that the bottleneck for autonomous scientific discovery is shifting from prescribing agent workflows to designing agent environments: the resources, constraints, and interfaces that shape agent behavior. We frame this as environment engineering: building environments that amplify productive behaviors, such as open-ended exploration, systematic artifact management, and inter-agent collaboration, while suppressing harmful behaviors, such as reward hacking and high-friction human oversight. We present EurekAgent, an environment-engineered agent system for metric-driven autonomous scientific discovery. EurekAgent engineers the environment along four dimensions: permissions engineering for bounded agent execution and isolated evaluation; artifact engineering for filesystem and Git-based collaboration; budget engineering for budget-aware exploration; and human-in-the-loop engineering for easy human supervision and intervention. EurekAgent sets new state-of-the-art results on multiple mathematics, kernel engineering, and machine learning tasks, including new state-of-the-art 26-circle packing results discovered with less than $11 in total API cost. We open-source our code and results, and call for environment engineering as a core research direction for developing reliable autonomous research agents.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2606.13662 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.13662 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.13662 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.