{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:LUOGLLASVK43SCRYXSJGT243GB","short_pith_number":"pith:LUOGLLAS","canonical_record":{"source":{"id":"2408.08435","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-15T21:59:23Z","cross_cats_sorted":[],"title_canon_sha256":"39006fd75159d83b2f62f511b693c2d1ab14e255800e1fb3331ee7bc06cba4be","abstract_canon_sha256":"8db6332235ff2e9fda484dbfc136fab44fc292aa1ed0294d556aa480bfa1d2b4"},"schema_version":"1.0"},"canonical_sha256":"5d1c65ac12aab9b90a38bc9269eb9b307b37bcdc4b21ec2649f9c1a6f9634677","source":{"kind":"arxiv","id":"2408.08435","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2408.08435","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"arxiv_version","alias_value":"2408.08435v2","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.08435","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"pith_short_12","alias_value":"LUOGLLASVK43","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"LUOGLLASVK43SCRY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"LUOGLLAS","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:LUOGLLASVK43SCRYXSJGT243GB","target":"record","payload":{"canonical_record":{"source":{"id":"2408.08435","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-15T21:59:23Z","cross_cats_sorted":[],"title_canon_sha256":"39006fd75159d83b2f62f511b693c2d1ab14e255800e1fb3331ee7bc06cba4be","abstract_canon_sha256":"8db6332235ff2e9fda484dbfc136fab44fc292aa1ed0294d556aa480bfa1d2b4"},"schema_version":"1.0"},"canonical_sha256":"5d1c65ac12aab9b90a38bc9269eb9b307b37bcdc4b21ec2649f9c1a6f9634677","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:53.065356Z","signature_b64":"aR48idS6pfSKTjeviKgBQ2OotIc/Cz2q+/Nsf8A0Ih8l17UKvbcdBCmR0SLA/XZnqLpaCWQYsVUAA+E8/uVfAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5d1c65ac12aab9b90a38bc9269eb9b307b37bcdc4b21ec2649f9c1a6f9634677","last_reissued_at":"2026-05-17T23:38:53.064734Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:53.064734Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2408.08435","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EwEN2XPTxRFBxJhX6vbNquyXZKEPwPRt1VShX3gDfTZCBmvQ7mjdKahG5k1I5JRx/XCD3MMSerzkyBkxTl5cDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T20:10:54.790088Z"},"content_sha256":"e40cf6f4b9e9116bdb2c6f96c559bd343c0bee76c092d8a0ee3e597c7b8ee6e2","schema_version":"1.0","event_id":"sha256:e40cf6f4b9e9116bdb2c6f96c559bd343c0bee76c092d8a0ee3e597c7b8ee6e2"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:LUOGLLASVK43SCRYXSJGT243GB","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Automated Design of Agentic Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Cong Lu, Jeff Clune, Shengran Hu","submitted_at":"2024-08-15T21:59:23Z","abstract_excerpt":"Researchers are investing substantial effort in developing powerful general-purpose agents, wherein Foundation Models are used as modules within agentic systems (e.g. Chain-of-Thought, Self-Reflection, Toolformer). However, the history of machine learning teaches us that hand-designed solutions are eventually replaced by learned solutions. We describe a newly forming research area, Automated Design of Agentic Systems (ADAS), which aims to automatically create powerful agentic system designs, including inventing novel building blocks and/or combining them in new ways. We further demonstrate tha"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Through extensive experiments across multiple domains including coding, science, and math, we show that our algorithm can progressively invent agents with novel designs that greatly outperform state-of-the-art hand-designed agents. Importantly, we consistently observe the surprising result that agents invented by Meta Agent Search maintain superior performance even when transferred across domains and models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The meta-agent can reliably generate functional, non-trivial new agent code from the archive without the search becoming dominated by ineffective or hallucinated programs, and that performance gains on the tested tasks reflect genuine improvements rather than overfitting to the evaluation setup.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c33b1871f07524ab761e9bffb9ce58b91d078ea51102f8ca6290f03df01bd9e3"},"source":{"id":"2408.08435","kind":"arxiv","version":2},"verdict":{"id":"f89bf277-13be-4d72-9192-fdebae510621","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T08:03:09.049437Z","strongest_claim":"Through extensive experiments across multiple domains including coding, science, and math, we show that our algorithm can progressively invent agents with novel designs that greatly outperform state-of-the-art hand-designed agents. Importantly, we consistently observe the surprising result that agents invented by Meta Agent Search maintain superior performance even when transferred across domains and models.","one_line_summary":"Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The meta-agent can reliably generate functional, non-trivial new agent code from the archive without the search becoming dominated by ineffective or hallucinated programs, and that performance gains on the tested tasks reflect genuine improvements rather than overfitting to the evaluation setup.","pith_extraction_headline":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains."},"references":{"count":208,"sample":[{"doi":"","year":null,"title":"Journal of Machine Learning Research , volume=","work_id":"624d20f4-9b7f-40ba-9536-7fdb2809375e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"International Conference on Machine Learning , pages=","work_id":"2a811334-3665-4f71-85af-e2a971e08c1d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"The Twelfth International Conference on Learning Representations , year=","work_id":"d8806a86-45c1-4d37-80a1-89ed680f9e17","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Autoagents: A framework for automatic agent generation","work_id":"6863e6b9-5143-4d8c-b7a2-276625d012f5","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Forty-first International Conference on Machine Learning , year=","work_id":"2f445f15-0b75-490e-b5b9-01572e15f1de","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":208,"snapshot_sha256":"af74acfc2b9c6af9ecf80826e6a249911e72aba2618ffadca9b060afb9e0e20e","internal_anchors":20},"formal_canon":{"evidence_count":2,"snapshot_sha256":"317b30ad6bb3c205d4d0758c2ea8d1cedbc085fcd34f1cdc74c79c8356a7940b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"f89bf277-13be-4d72-9192-fdebae510621"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0MlJx2jzXe2Hu+toBfdsEeB9CR+h0KLFl8AtX30wdJGCiogs+2Utfr0QRTnmBWHD5Qch94iLWa5hoAQGI+NjBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T20:10:54.790628Z"},"content_sha256":"d47427e91e20e329518a37335bd2983bbcef0ca097cb99271e60b058bb598fe9","schema_version":"1.0","event_id":"sha256:d47427e91e20e329518a37335bd2983bbcef0ca097cb99271e60b058bb598fe9"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/LUOGLLASVK43SCRYXSJGT243GB/bundle.json","state_url":"https://pith.science/pith/LUOGLLASVK43SCRYXSJGT243GB/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/LUOGLLASVK43SCRYXSJGT243GB/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-03T20:10:54Z","links":{"resolver":"https://pith.science/pith/LUOGLLASVK43SCRYXSJGT243GB","bundle":"https://pith.science/pith/LUOGLLASVK43SCRYXSJGT243GB/bundle.json","state":"https://pith.science/pith/LUOGLLASVK43SCRYXSJGT243GB/state.json","well_known_bundle":"https://pith.science/.well-known/pith/LUOGLLASVK43SCRYXSJGT243GB/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:LUOGLLASVK43SCRYXSJGT243GB","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"8db6332235ff2e9fda484dbfc136fab44fc292aa1ed0294d556aa480bfa1d2b4","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-15T21:59:23Z","title_canon_sha256":"39006fd75159d83b2f62f511b693c2d1ab14e255800e1fb3331ee7bc06cba4be"},"schema_version":"1.0","source":{"id":"2408.08435","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2408.08435","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"arxiv_version","alias_value":"2408.08435v2","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.08435","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"pith_short_12","alias_value":"LUOGLLASVK43","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"LUOGLLASVK43SCRY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"LUOGLLAS","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:d47427e91e20e329518a37335bd2983bbcef0ca097cb99271e60b058bb598fe9","target":"graph","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Through extensive experiments across multiple domains including coding, science, and math, we show that our algorithm can progressively invent agents with novel designs that greatly outperform state-of-the-art hand-designed agents. Importantly, we consistently observe the surprising result that agents invented by Meta Agent Search maintain superior performance even when transferred across domains and models."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The meta-agent can reliably generate functional, non-trivial new agent code from the archive without the search becoming dominated by ineffective or hallucinated programs, and that performance gains on the tested tasks reflect genuine improvements rather than overfitting to the evaluation setup."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains."}],"snapshot_sha256":"c33b1871f07524ab761e9bffb9ce58b91d078ea51102f8ca6290f03df01bd9e3"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"317b30ad6bb3c205d4d0758c2ea8d1cedbc085fcd34f1cdc74c79c8356a7940b"},"paper":{"abstract_excerpt":"Researchers are investing substantial effort in developing powerful general-purpose agents, wherein Foundation Models are used as modules within agentic systems (e.g. Chain-of-Thought, Self-Reflection, Toolformer). However, the history of machine learning teaches us that hand-designed solutions are eventually replaced by learned solutions. We describe a newly forming research area, Automated Design of Agentic Systems (ADAS), which aims to automatically create powerful agentic system designs, including inventing novel building blocks and/or combining them in new ways. We further demonstrate tha","authors_text":"Cong Lu, Jeff Clune, Shengran Hu","cross_cats":[],"headline":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-15T21:59:23Z","title":"Automated Design of Agentic Systems"},"references":{"count":208,"internal_anchors":20,"resolved_work":208,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Journal of Machine Learning Research , volume=","work_id":"624d20f4-9b7f-40ba-9536-7fdb2809375e","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"International Conference on Machine Learning , pages=","work_id":"2a811334-3665-4f71-85af-e2a971e08c1d","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"The Twelfth International Conference on Learning Representations , year=","work_id":"d8806a86-45c1-4d37-80a1-89ed680f9e17","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":6,"title":"Autoagents: A framework for automatic agent generation","work_id":"6863e6b9-5143-4d8c-b7a2-276625d012f5","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":7,"title":"Forty-first International Conference on Machine Learning , year=","work_id":"2f445f15-0b75-490e-b5b9-01572e15f1de","year":null}],"snapshot_sha256":"af74acfc2b9c6af9ecf80826e6a249911e72aba2618ffadca9b060afb9e0e20e"},"source":{"id":"2408.08435","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T08:03:09.049437Z","id":"f89bf277-13be-4d72-9192-fdebae510621","model_set":{"reader":"grok-4.3"},"one_line_summary":"Meta Agent Search uses a meta-agent to iteratively program novel agentic systems in code, producing agents that outperform state-of-the-art hand-designed ones across coding, science, and math while transferring across domains and models.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A meta-agent can program new agents in code that outperform hand-designed ones and transfer across domains.","strongest_claim":"Through extensive experiments across multiple domains including coding, science, and math, we show that our algorithm can progressively invent agents with novel designs that greatly outperform state-of-the-art hand-designed agents. Importantly, we consistently observe the surprising result that agents invented by Meta Agent Search maintain superior performance even when transferred across domains and models.","weakest_assumption":"The meta-agent can reliably generate functional, non-trivial new agent code from the archive without the search becoming dominated by ineffective or hallucinated programs, and that performance gains on the tested tasks reflect genuine improvements rather than overfitting to the evaluation setup."}},"verdict_id":"f89bf277-13be-4d72-9192-fdebae510621"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:e40cf6f4b9e9116bdb2c6f96c559bd343c0bee76c092d8a0ee3e597c7b8ee6e2","target":"record","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"8db6332235ff2e9fda484dbfc136fab44fc292aa1ed0294d556aa480bfa1d2b4","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-08-15T21:59:23Z","title_canon_sha256":"39006fd75159d83b2f62f511b693c2d1ab14e255800e1fb3331ee7bc06cba4be"},"schema_version":"1.0","source":{"id":"2408.08435","kind":"arxiv","version":2}},"canonical_sha256":"5d1c65ac12aab9b90a38bc9269eb9b307b37bcdc4b21ec2649f9c1a6f9634677","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5d1c65ac12aab9b90a38bc9269eb9b307b37bcdc4b21ec2649f9c1a6f9634677","first_computed_at":"2026-05-17T23:38:53.064734Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:53.064734Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"aR48idS6pfSKTjeviKgBQ2OotIc/Cz2q+/Nsf8A0Ih8l17UKvbcdBCmR0SLA/XZnqLpaCWQYsVUAA+E8/uVfAA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:53.065356Z","signed_message":"canonical_sha256_bytes"},"source_id":"2408.08435","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:e40cf6f4b9e9116bdb2c6f96c559bd343c0bee76c092d8a0ee3e597c7b8ee6e2","sha256:d47427e91e20e329518a37335bd2983bbcef0ca097cb99271e60b058bb598fe9"],"state_sha256":"902c42d1431317c0e4e027b7048cfbba3d5403bcaf7f5860d066cd6ac21fdd7c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RMPUAUTNG2rV56UdKxoFSr53KowXW3BPyGos5eIPvNoJVoEQq+zNLGBqNZu9ZbDSa8bRRlj0hdMigLH+LlfeBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-03T20:10:54.793111Z","bundle_sha256":"c8ad5ccaef70fe70177518f32a9d4cfc0eb4470c90ba660f2d89c11ccdd2a0d9"}}