在多租户 SaaS 环境中扩展智能体 AI.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.I S V 3 0 9Scaling agentic AI in multi-tenant SaaS environmentsHrushi Gangur(He/Him)Principal Solutions ArchitectRielah De Jesus(She/Her)Principal Solutions Architec

2、t 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.F R O M P O C T O P R O D U C T I O N C Y C L EThe 9-month trapThe reality check:Week 1-2:”AI agents work great in our demo!”Month 3:”Wait,our enterprise customers have requirements.”Month 6:”Every tenant has different model prefer

3、ence and tools integrations”Month 9:Were rebuilding everything.againTwo critical multi-tenant blockers:1.Backend model choices-One size doesnt fit all tenants2.3P tool integrations-Every tenant has a different ISV stack 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AI agentTools

4、AgentUserAppTask ResponseAgentModelsC O N C E P T SMemory and Context Management 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Agentic AI systemAgent instructionAgent local toolsAgent contextFrameworkAppModelsAgentAgentC O N C E P T SMemory and context managementToolsThe LLM cho

5、ice and tool integration arent just featurestheyre the foundation that everything else builds on.MCPA2A 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.From the brain to the handsThe agentic workflow:Models:Give agents reasoning abilityTools:Give agents ability to act In the real

6、worldM O D E L O P T I O N A L I T Y 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Multi-tenancy driversM O D E L O P T I O N A L I T YSecurity perimeter“Our data cannot leave agents application boundary”Infrastructure alignmentPerformance vs cost trade-offsGlobal constraintsSpe