AWS re:Invent 2025 - Build agentic workflows on AWS with third-party agents and tools (AIM3311)

Building Agentic Workflows on AWS with Third-Party Agents and Tools

Overview of AWS Agentic Stack

• AWS provides a comprehensive "agentic stack" for building intelligent agent-based applications
• Key components include:
  • Core services like Amazon Bedrock, Agent Core Runtime, Agent Core Gateway, and Agent Core Observability
  • Architectural principles and recommendations based on customer feedback
  • Integration with third-party AI agents and tools available on AWS Marketplace

AWS Marketplace for AI Agents and Tools

• AWS Marketplace provides an enterprise-ready digital catalog of software, services, and AI agents/tools
• Key benefits:
  • Reduced procurement time (from months to minutes)
  • Unified billing and cost control transparency
  • Simplified deployment through native AWS service integrations
• Three main categories of offerings:
  1. API-based agent tools (remote, ISV-hosted)
  2. Container-based agents and tools (designed to run on Agent Core Runtime)
  3. Agentic applications and professional services

Workday's Experience Building a Planning Agent

• Workday has an agent platform team that provides a common stack and consistency for building agents
• The Planning Agent represents 4 personas: analyst, modeler, planner, and admin
• Key challenges and lessons learned:
  • Initial approach of sending data to LLMs faced issues with token limits and accuracy
  • Chunking data and sending to LLM improved, but introduced latency and cost concerns
  • Ultimately, leveraging LLMs to generate code for data analysis proved most effective
  • Importance of a secure, sandboxed code interpreter (provided by Agent Core)
• Results:
  • 50% reduction in tokens sent, improved accuracy, and scalable solution
  • Rapid time-to-production (3 days) enabled by Agent Core capabilities

Building an Agentic Workflow with Third-Party Tools

• Demonstrated a "Company Research Agent" use case for sales teams
• Key steps:
  1. Architect the solution components (user, executor, LLM, tools)
  2. Set up the tools (Bedrock model, Tavil search tool) in the cloud environment
  3. Implement the code using LangChain framework
  • Leverage LLM (Anthropic Claude 3.5 HiQ)
  • Define prompts and tools (web search, local document search)
  • Orchestrate the agent execution
• Emphasized the importance of security, governance, performance, reliability, and cost control when moving to production

Deploying a Pre-Built Agent from AWS Marketplace

• Demonstrated deploying a pre-built "CRM AI Sales Agent" from AWS Marketplace
• Key steps:
  1. Discover the agent based on business needs
  2. Subscribe and accept vendor terms in a single click
  3. Deploy the agent on Agent Core Runtime
• Benefits:
  • Inherit vendor's work on prompt engineering, LLM selection, and component integration
  • Run securely within customer's VPC with IAM-based access control
  • Minimal setup required (just provide environment variables)

Key Takeaways

• AWS provides a comprehensive "agentic stack" of services and tools to build intelligent agent-based applications
• AWS Marketplace offers a growing ecosystem of pre-built AI agents and tools to accelerate development
• Workday's experience highlights the importance of secure, sandboxed code execution and iterating on approaches to leverage LLMs effectively
• Building agentic workflows involves architecting the right components (LLM, tools, prompts) and addressing production-readiness concerns
• Pre-built agents from AWS Marketplace can provide a fast path to deployment with inherent security and governance