{"id":"https://openalex.org/W6910824972","doi":"https://doi.org/10.48550/arxiv.2503.13657","title":"Why Do Multi-Agent LLM Systems Fail?","display_name":"Why Do Multi-Agent LLM Systems Fail?","publication_year":2025,"publication_date":"2025-03-17","ids":{"openalex":"https://openalex.org/W6910824972","doi":"https://doi.org/10.48550/arxiv.2503.13657"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2503.13657","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.13657","pdf_url":"https://arxiv.org/pdf/2503.13657","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2503.13657","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5028114931","display_name":"Mert Cemri","orcid":"https://orcid.org/0000-0003-1515-6713","observed_orcids":["https://orcid.org/0000-0003-1515-6713"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cemri, Mert","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126852859","display_name":"Melissa Pan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pan, Melissa Z.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Yang, Shuyi","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Shuyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308487","display_name":"Lakshya A Agrawal","orcid":"https://orcid.org/0000-0003-0409-8212","observed_orcids":["https://orcid.org/0000-0003-0409-8212","https://orcid.org/0009-0001-5729-3903"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Agrawal, Lakshya A.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092090328","display_name":"Bhavya Chopra","orcid":"https://orcid.org/0000-0001-8638-3863","observed_orcids":["https://orcid.org/0000-0001-8638-3863"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chopra, Bhavya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126942968","display_name":"Rishabh Tiwari","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tiwari, Rishabh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Keutzer, Kurt","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Keutzer, Kurt","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013608601","display_name":"Aditya Parameswaran","orcid":"https://orcid.org/0000-0002-4538-4752","observed_orcids":["https://orcid.org/0000-0002-4538-4752"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Parameswaran, Aditya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004921249","display_name":"Dan Klein","orcid":"https://orcid.org/0000-0002-8881-1902","observed_orcids":["https://orcid.org/0000-0002-8881-1902","https://orcid.org/0009-0002-5983-2904"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Klein, Dan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030620564","display_name":"Kannan Ramchandran","orcid":"https://orcid.org/0000-0002-4567-328X","observed_orcids":["https://orcid.org/0000-0002-4567-328X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ramchandran, Kannan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005554337","display_name":"Matei Zaharia","orcid":"https://orcid.org/0000-0002-7547-7204","observed_orcids":["https://orcid.org/0000-0002-7547-7204"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zaharia, Matei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Gonzalez, Joseph E.","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gonzalez, Joseph E.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Stoica, Ion","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Stoica, Ion","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":21,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":true,"primary_topic":{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.8823000192642212,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.8823000192642212,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.02459999918937683,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T13904","display_name":"Artificial Intelligence Applications","score":0.010200000368058681,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/multi-agent-llm-systems","display_name":"multi-agent LLM systems","score":0.9610000252723694},{"id":"https://openalex.org/keywords/failure-analysis","display_name":"failure analysis","score":0.6439999938011169},{"id":"https://openalex.org/keywords/failure-taxonomy","display_name":"failure taxonomy","score":0.8220000267028809},{"id":"https://openalex.org/keywords/llm-as-a-judge","display_name":"LLM-as-a-judge","score":0.8320000171661377}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7400000095367432},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.4717999994754791},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.460999995470047},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.42730000615119934},{"id":"https://openalex.org/C2781354396","wikidata":"https://www.wikidata.org/wiki/Q840407","display_name":"Enthusiasm","level":2,"score":0.4156000018119812},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.40619999170303345},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.38839998841285706},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.33799999952316284},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.32690000534057617},{"id":"https://openalex.org/C169087156","wikidata":"https://www.wikidata.org/wiki/Q2131593","display_name":"Framing (construction)","level":2,"score":0.32440000772476196},{"id":"https://openalex.org/C58642233","wikidata":"https://www.wikidata.org/wiki/Q8269924","display_name":"Taxonomy (biology)","level":2,"score":0.32350000739097595},{"id":"https://openalex.org/C143587482","wikidata":"https://www.wikidata.org/wiki/Q1543216","display_name":"Iterative and incremental development","level":2,"score":0.3019999861717224},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2996000051498413},{"id":"https://openalex.org/C115903868","wikidata":"https://www.wikidata.org/wiki/Q80993","display_name":"Software engineering","level":1,"score":0.26910001039505005}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2503.13657","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.13657","pdf_url":"https://arxiv.org/pdf/2503.13657","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2503.13657","doi":"https://doi.org/10.48550/arxiv.2503.13657","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.13657","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2503.13657","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.13657","pdf_url":"https://arxiv.org/pdf/2503.13657","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7164642796","https://openalex.org/W4416255023","https://openalex.org/W7140346934","https://openalex.org/W7141713967","https://openalex.org/W7131436133","https://openalex.org/W7161680277","https://openalex.org/W7163007909","https://openalex.org/W7167897761","https://openalex.org/W7125816407","https://openalex.org/W6948524426"],"abstract_inverted_index":{"Despite":[0],"enthusiasm":[1],"for":[2,22,71,85,201],"Multi-Agent":[3,91],"LLM":[4,215],"Systems":[5],"(MAS),":[6],"their":[7],"performance":[8],"gains":[9],"on":[10],"popular":[11,54],"benchmarks":[12],"are":[13],"often":[14],"minimal.":[15],"This":[16,120],"gap":[17],"highlights":[18],"a":[19,23,44,198],"critical":[20],"need":[21],"principled":[24],"understanding":[25],"of":[26,38,47,75,83,102],"why":[27],"MAS":[28,55,70,183],"fail.":[29],"Addressing":[30],"this":[31],"question":[32],"requires":[33],"systematic":[34,81],"identification":[35],"and":[36,111,137,159,172,213,221],"analysis":[37,101,186],"failure":[39,67,163],"patterns.":[40],"We":[41,96,156,204],"introduce":[42],"MAST-Data,":[43,86],"comprehensive":[45,208],"dataset":[46,63,209],"1600+":[48],"annotated":[49],"traces":[50],"collected":[51],"across":[52,165],"7":[53],"frameworks.":[56],"MAST-Data":[57,160],"is":[58],"the":[59,66,73,89,211],"first":[60,90],"multi-agent":[61],"system":[62,131],"to":[64,161,217],"outline":[65],"dynamics":[68],"in":[69,223],"guiding":[72],"development":[74,222],"better":[76,182],"future":[77,202],"systems.":[78],"To":[79,141],"enable":[80,142],"classification":[82],"failures":[84,192],"we":[87,145],"build":[88],"System":[92],"Failure":[93],"Taxonomy":[94],"(MAST).":[95],"develop":[97,146],"MAST":[98,158],"through":[99],"rigorous":[100],"150":[103],"traces,":[104],"guided":[105],"closely":[106],"by":[107,113],"expert":[108],"human":[109,154],"annotators":[110],"validated":[112],"high":[114,151],"inter-annotator":[115],"agreement":[116,152],"(kappa":[117],"=":[118],"0.88).":[119],"process":[121],"identifies":[122],"14":[123],"unique":[124],"modes,":[125],"clustered":[126],"into":[127],"3":[128],"categories:":[129],"(i)":[130],"design":[132],"issues,":[133],"(ii)":[134],"inter-agent":[135],"misalignment,":[136],"(iii)":[138],"task":[139],"verification.":[140],"scalable":[143],"annotation,":[144],"an":[147],"LLM-as-a-Judge":[148],"pipeline":[149],"with":[150,153],"annotations.":[155],"leverage":[157],"analyze":[162],"patterns":[164],"models":[166],"(GPT4,":[167],"Claude":[168],"3,":[169],"Qwen2.5,":[170],"CodeLlama)":[171],"tasks":[173],"(coding,":[174],"math,":[175],"general":[176],"agent),":[177],"demonstrating":[178],"improvement":[179],"headrooms":[180],"from":[181],"design.":[184],"Our":[185],"provides":[187],"insights":[188],"revealing":[189],"that":[190],"identified":[191],"require":[193],"more":[194],"sophisticated":[195],"solutions,":[196],"highlighting":[197],"clear":[199],"roadmap":[200],"research.":[203],"publicly":[205],"release":[206],"our":[207,214],"(MAST-Data),":[210],"MAST,":[212],"annotator":[216],"facilitate":[218],"widespread":[219],"research":[220],"MAS.":[224]},"counts_by_year":[{"year":2026,"cited_by_count":14},{"year":2025,"cited_by_count":7}],"updated_date":"2026-10-31T12:09:11.096373","created_date":"2025-10-10T00:00:00"}