{"paper":{"title":"AI Harness Engineering: A Runtime Substrate for Foundation-Model Software Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Software-engineering capability for foundation-model agents emerges from a model-harness-environment system rather than from the model alone.","cross_cats":["cs.AI"],"primary_cat":"cs.SE","authors_text":"Hailin Zhong, Shengxin Zhu","submitted_at":"2026-05-13T11:14:59Z","abstract_excerpt":"Foundation models have transformed automated code generation, yet autonomous software-engineering agents remain unreliable in realistic development settings. The dominant explanation locates this gap in model capability. We propose a different locus: software-engineering capability emerges from a model-harness-environment system, in which a runtime substrate -- the harness -- mediates how a foundation-model agent observes a project, acts on it, receives feedback, and establishes that a change is complete. We formalize this substrate as an AI Harness Engineering and identify eleven component re"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose a different locus: software-engineering capability emerges from a model-harness-environment system, in which a runtime substrate -- the harness -- mediates how a foundation-model agent observes a project, acts on it, receives feedback, and establishes that a change is complete.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That defining eleven component responsibilities and a four-level ladder will systematically improve agent reliability and verifiability, an assumption stated without supporting data or derivation in the abstract.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"The paper defines AI Harness Engineering as a runtime substrate with eleven components and a four-level ladder that reframes agent reliability as a model-harness-environment system property rather than model capability alone.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Software-engineering capability for foundation-model agents emerges from a model-harness-environment system rather than from the model alone.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"38745aeb6cb7ad7e0642a2fb3dedd7eacb6f398269ad58f160f8f180eb1fa2d2"},"source":{"id":"2605.13357","kind":"arxiv","version":1},"verdict":{"id":"f734b0c7-8b0d-4bdd-9d7c-aac713be3f00","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:10:33.629110Z","strongest_claim":"We propose a different locus: software-engineering capability emerges from a model-harness-environment system, in which a runtime substrate -- the harness -- mediates how a foundation-model agent observes a project, acts on it, receives feedback, and establishes that a change is complete.","one_line_summary":"The paper defines AI Harness Engineering as a runtime substrate with eleven components and a four-level ladder that reframes agent reliability as a model-harness-environment system property rather than model capability alone.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That defining eleven component responsibilities and a four-level ladder will systematically improve agent reliability and verifiability, an assumption stated without supporting data or derivation in the abstract.","pith_extraction_headline":"Software-engineering capability for foundation-model agents emerges from a model-harness-environment system rather than from the model alone."},"references":{"count":18,"sample":[{"doi":"","year":2021,"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","ref_index":1,"cited_arxiv_id":"2107.03374","is_internal_anchor":true},{"doi":"","year":2020,"title":"Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, et al","work_id":"857e276a-33ad-44d4-9d43-c728bd2db249","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan","work_id":"faab4af1-0cb8-4520-9a0a-c7a1a55118e4","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press","work_id":"7a8ddf3e-1730-4670-9cf2-8789e3706977","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang H","work_id":"ba783dfe-165d-4d63-b7f9-9708af7040b3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":18,"snapshot_sha256":"92dab86cc85b5856542c23b8222e918e76631680ad80ee1ff237fcbb35bbd41b","internal_anchors":2},"formal_canon":{"evidence_count":1,"snapshot_sha256":"f4ea6edfd57c9f674ea0a01065b5e807f2e7ea030a40fc70edb6d925cb813dfa"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}