{"paper":{"title":"AgentDisCo: Towards Disentanglement and Collaboration in Open-ended Deep Research Agents","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"AgentDisCo separates exploration and exploitation into critic and generator agents that iteratively refine outlines before a report writer synthesizes them.","cross_cats":["cs.CL","cs.MA","cs.MM"],"primary_cat":"cs.IR","authors_text":"Jiarui Jin, Shijian Wang, Wenxiang Jiao, Yuan Lu, Zexuan Yan","submitted_at":"2026-05-12T08:14:15Z","abstract_excerpt":"In this paper, we present AgentDisCo, a novel Disentangled and Collaborative agentic architecture that formulates deep research as an adversarial optimization problem between information exploration and exploitation. Unlike existing approaches that conflate these two processes into a single module, AgentDisCo employs a critic agent to evaluate generated outlines and refine search queries, and a generator agent to retrieve updated results and revise outlines accordingly. The iteratively refined outline is then passed to a downstream report writer that synthesizes a comprehensive research report"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"AgentDisCo employs a critic agent to evaluate generated outlines and refine search queries, and a generator agent to retrieve updated results and revise outlines accordingly. The iteratively refined outline is then passed to a downstream report writer that synthesizes a comprehensive research report.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That formulating deep research as an adversarial optimization problem between information exploration and exploitation via separate critic and generator agents yields superior results compared to conflated single-module approaches, and that the meta-optimization harness can reliably discover effective strategies without extensive human intervention.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AgentDisCo disentangles exploration and exploitation in research agents via critic-generator collaboration and meta-optimization, achieving competitive performance on benchmarks while introducing the GALA benchmark from user browsing histories.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AgentDisCo separates exploration and exploitation into critic and generator agents that iteratively refine outlines before a report writer synthesizes them.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3c133b4a6ae4f39159c0686ae656823a5b2bcec4230dcf8cb0ddd94309515d1b"},"source":{"id":"2605.11732","kind":"arxiv","version":2},"verdict":{"id":"ace29957-fb6d-4737-8c73-c94a1fd51c06","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T05:33:14.507012Z","strongest_claim":"AgentDisCo employs a critic agent to evaluate generated outlines and refine search queries, and a generator agent to retrieve updated results and revise outlines accordingly. The iteratively refined outline is then passed to a downstream report writer that synthesizes a comprehensive research report.","one_line_summary":"AgentDisCo disentangles exploration and exploitation in research agents via critic-generator collaboration and meta-optimization, achieving competitive performance on benchmarks while introducing the GALA benchmark from user browsing histories.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That formulating deep research as an adversarial optimization problem between information exploration and exploitation via separate critic and generator agents yields superior results compared to conflated single-module approaches, and that the meta-optimization harness can reliably discover effective strategies without extensive human intervention.","pith_extraction_headline":"AgentDisCo separates exploration and exploitation into critic and generator agents that iteratively refine outlines before a report writer synthesizes them."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11732/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-26T20:47:04.925796Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T23:31:31.895735Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-20T13:47:05.575242Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-20T03:42:00.421420Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"69e338cf1fd974abf4a0eb19e65d6589b4ef45608831b835dd4fafc8a265f227"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"db6b8cde9abc7ca606f81be559b7c8641bf90294e1df54ab32d73df9f6e0a083"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}