SUMM

Building Agents with Amazon Nova Act and MCP - Du'An Lightfoot, Amazon (Full Workshop)

Introduction to Agents and AWS 00:23

  • Dewan Lightfoot introduces the workshop focused on building intelligent autonomous AI systems using Amazon Nova ACT and MCP.
  • He emphasizes the significance of agentic AI systems in enhancing applications and businesses.

Understanding Agentic Systems 01:35

  • Key features of agentic systems include planning, action-taking, and reasoning capabilities.
  • The architecture consists of user input, agentic systems, potential human oversight, and generated responses.
  • Continuous evaluation is crucial for optimizing prompts and assessing system performance.

Use Cases for Agentic Systems 04:04

  • Agentic systems are ideal for complex tasks requiring multiple tools, while simple tasks can be efficiently handled with traditional solutions.

Approaches to Building Agents on AWS 04:45

  • Three approaches to consider: specialized use with Amazon Q, fully managed agents with Amazon Bedrock, and DIY options using Strands agents.
  • Strands agents are lightweight and open source, allowing for easy implementation in fewer lines of code.

Features of Amazon Nova ACT 07:22

  • Nova ACT, a research preview model, enables complex tasks like browsing the internet for research and interacting with Amazon.com.
  • The integration with the Modern Context Protocol (MCP) enhances functionality and workflow management.

Workshop Setup and Modules 09:42

  • Participants engage in a hands-on workshop where they set up AWS accounts and access models.
  • The workshop consists of three modules: getting started with Nova ACT, creating an MCP server, and using Strands agents.

Utilizing Amazon Bedrock and Models 12:14

  • Bedrock provides serverless API access to various foundation models, allowing for extensive generative applications.
  • Participants learn to enable specific models for the workshop.

Implementing Nova ACT 16:06

  • Nova ACT is demonstrated with simple commands to search for products, showcasing its ease of use compared to traditional web automation methods.

Limitations and Challenges 22:10

  • The system struggles with CAPTCHAs and may get stuck in loops, highlighting the importance of careful prompt design.

Multi-Agent Collaboration with Strands 43:11

  • Strands allows for creating multiple agents, each with specific roles and tools, facilitating complex workflows like cloud migration planning.
  • An orchestrator agent coordinates tasks among various specialized agents.

Conclusion and Next Steps 54:46

  • Lightfoot emphasizes the potential of agentic workflows and encourages participants to explore and build with the tools demonstrated.
  • A survey is introduced for feedback, with AWS credits available for completion.
HomeSubmitSaved