Strands Agents

Ref: https://www.udemy.com/course/ultimate-aws-certified-generative-ai-developer-professional/learn/lecture/53684339 and https://aws.amazon.com/blogs/opensource/introducing-strands-agents-an-open-source-ai-agents-sdk/

• 🔧 Open-source Python SDK for Agentic AI

  • Made by Amazon
  • Can create “agent harnesses”, specialized agents, and multi-agent systems (complex task decomposition, collaborative task solving…)
    • 💡 Other similar SDKs that allow creating “agent harnesses”: LangGraph, OpenAI Agents SDK, Google ADK…

  

• Features

  • Multimodal support (text, audio, images…)
  • MCP support
  • Better AWS integrations than competitors via Boto3 library (e.g. seamless integration with Bedrock FMs, Lambda…)
  • Support for non-AWS tools and models (e.g. OpenAI models)
  • Many built-in tools: http support for web search, run shell commands, speak (via Amazon Polly), Mem0 agent memory & personalization… →

• Strands Agent Loop: loop between LLM reasoning, tool selection, and tool execution until final response is satisfactory

• 💡 Quick start guide: https://strandsagents.com/docs/user-guide/quickstart/python/

AWS Agent Squad

Ref: https://www.udemy.com/course/ultimate-aws-certified-generative-ai-developer-professional/learn/lecture/53684345

• 🔧 Open-source framework for multi-agent workflows, focused on intelligent intent classification
  • Route queries to the right agent
  • Created by AWS, now maintained by 2FastLabs: https://github.com/2FastLabs/agent-squad
• High-level architecture flow diagram
• Features
  • Shared contexts across agents
  • Conversation memory & orchestration
  • Many pre-built agents and classifiers
  • Good integration with AWS products (Bedrock Agents, extend Bedrock Flows…)
  • Supervisor Agent for multi-agent coordination
• ‼️ Main difference between Strands Agents and Agent Squad?
  • Strands Agents focused on tool use in a single agent loop
  • Agent Squad focused on routing/intent detection

Amazon Bedrock AgentCore

AgentCore - Introduction

Ref: https://www.udemy.com/course/ultimate-aws-certified-generative-ai-developer-professional/learn/lecture/53684351

• If you build an AI agent with an Agent SDK (e.g. LangGraph, Strands Agents SDK, OpenAI Agents SDK…), you then have to handle its deployment, which can get complex

  • Deploying locally? Must for example handling containers…
  • …But you can deploy the agent with managed solutions in the cloud, e.g. Bedrock AgentCore!

• 🔧 AgentCore offers serverless deployment & operation of AI agents at scale

  • Includes additional tools & capabilities for your agents
  • Includes “starter toolkit” → very easy to deploy agents to AWS

  

• Support for ANY agent framework

  • OpenAI Agents SDK, Google ADK, LangGraph, CrewAI…
  • Can handle Bedrock Agents configured in Bedrock
  • Great support for agents built with Strands Agents SDK