多智能体工作流失败的 10 个原因以及应对方法.pdf

1、110 Reasons Your Multi-Agent Workflows FailVictor Dibia,PhD|vykhurNov 18,2024QCON SF 2024And what you can do about itWhy are agents/Multi-Agent Systems?Interesting?23Imagine a scenario where computers could handle increasingly complex tasks on your behalf.4“Download email attachments from clients,lo

2、ad them into Excel.4Back office data entry across multiple systems 5“Build an Android app that can help users view and purchase stocks5Software engineering6“File my taxes6Finance77Back OfficeSoftware EngineeringFinance-Tedious and repetitive -Important -Involves many,sometimes proactive steps 83 Key

3、 Insights8-Save time(autonomous task completion)-A new digital interface-Disrupt current approaches to solving tasksOn why agents/multi-agent systems are so interesting right now!Forbes469%increase in#of agent startups YC)over the last 2 yearsCount of YC companies that explicitly mention AI Agents i

4、n their company description10Source:YC data11There seems to be universal agreement that.The future is Agentic12But.there are a few issues12Source:langchainSource:langchainAutonomous Agents have the last mile problemSource:Richard Socher CEO,Y)16While Agents are exciting,the Agents are sometimes Unre

5、liable!17A paradox and questions.-What are multi-agent systems and how to build them?-What factors drive reliability issues?-Should I invest in an autonomous multi-agent system?Talk Agenda What are autonomous multi-agent systems?How can you build them?Part 1IntroductionPart 2Failure ModesPart 3What

6、you can do1810 reasons current multi-agent workflows fail Key takeaways Next stepsVictor DibiaCore Contributor AutoGen,AutoGen StudioLeading OSS framework for building multi-agent applications.Previously Worked-Cloudera-ML Engineer-IBM Research-Research Staff Member 19Principal RSDE,Microsoft Resear