《构建人机协同的受控智能体 [重复].pdf》由会员分享,可在线阅读,更多相关《构建人机协同的受控智能体 [重复].pdf(29页珍藏版)》请在三个皮匠报告上搜索。
1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 9 7-RBuilding governed agents with Human-in-the-loopIsaac PriviteraPrincipal Data Scientist,Generative AI Innovation Center,AWSNick AldrigePrincipal Engineer
2、,Agentic AI,AWSCore Maintainer,MCPSteering Committee Member,A2A 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaAgents and AutonomyDesigning for ComplianceAgentCore PolicyActions and resources 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web
3、 Services,Inc.or its affiliates.All rights reserved.Agents and Autonomy 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Agentic AI maturity scale from rules to reasoningLow agencyMore human oversightHigh agencyLess human oversightRule-based RPAExecutes predefined rulesGenerative A
4、I assistantsAchieves a specific,predefined goalGoal-driven AI agentsOperates toward a higher-level objective,not just a taskFully autonomous agentic systemsIndependently sets and executes goals 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Agentic Loop Strands Agents ReasoningAc
5、tionsProcess the input using a language modelReceive a user inputDecide to use tools to gather information or perform actionsContinue reasoning with new informationGenerate a final responseExecute the tools and receive resultsEnables the agent to perform complex,multi-step reasoning,and actions 2025
6、,Amazon Web Services,Inc.or its affiliates.All rights reserved.7Finding a balance in building agentic AI systemsE M E R G I N G R I S K S A N D C H A L L E N G E SE X P E C T E D B E N E F I T S 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its af