Course Objective & Scope
Heavyweight orchestration frameworks often get in the way of modern frontier models that already excel at native reasoning, planning and parallel tool invocation. This programme teaches how to build clean, production-ready, model-driven agents in Python using the open-source AWS Strands Agents SDK.
The core agentic loop, custom tools with the @tool decorator, native Model Context Protocol (MCP) integration, switching across Amazon Bedrock, Anthropic, OpenAI and LiteLLM/Ollama providers, multi-agent patterns (Agents-as-Tools, Swarm, Workflow Graphs) and OpenTelemetry tracing are covered hands-on. Participants build and test a reusable Strands agent tailored to an internal use case.
Who Should Attend?
- AI and software engineers who want a lightweight, Pythonic SDK for production agents
- Cloud developers building on Amazon Bedrock and open-source tooling
- Technical teams decomposing complex workflows into specialized multi-agent systems
- Prerequisite: intermediate Python programming (type hints, async/await) and basic LLM API experience
Key Takeaways
- Configure a model-driven agent loop, system prompt and streaming response handler with Strands Agents SDK
- Author custom Python tools via
@tooland connect external MCP servers with minimal boilerplate - Compose multi-agent architectures using Agents-as-Tools, dynamic Swarms and deterministic Graphs
- Manage conversation history, session persistence, structured outputs and lifecycle hooks/guardrails
- Instrument agents with OpenTelemetry and deploy them to AWS Lambda, ECS/EKS or Bedrock AgentCore
Curriculum
The Model-Driven Approach and Strands Agents Fundamentals
- Moving from rigid chains to model-driven plan-act-reflect agent loops
- Configuring model providers (Bedrock, Anthropic, OpenAI, LiteLLM, Ollama) and Pydantic structured outputs
Tool Authoring (@tool) and Native MCP Integration
- Turning Python functions into schema-validated tools with @tool; async and parallel execution
- Connecting MCP servers over stdio, SSE and Streamable HTTP
State Management, Memory, Hooks and Guardrails
- Sliding-window and summarizing conversation managers plus session persistence
- Lifecycle hooks/callbacks for tool confirmation, auditing and Bedrock Guardrails integration
Multi-Agent Patterns: Agents-as-Tools, Swarms and Graphs
- Wrapping specialist agents as callable tools under a supervisor agent
- Autonomous peer handoffs in Swarms versus explicit topology control with Agent Graphs
Observability, Evaluation and Production Deployment
- Native OpenTelemetry traces and metrics exported to CloudWatch, Langfuse or Jaeger
- Trajectory evaluation and deploying onto Bedrock AgentCore, AWS Fargate and Lambda
Other Courses in This Category
-
Managed enterprise RAG with Amazon Bedrock Knowledge Bases (advanced parsing, chunking, hybrid search, reranking) and safety enforcement with Bedrock Guardrails (PII, grounding, content filters).
- Live Course
- 5 modules
-
Training, deploying and governing classical ML models and open-weight LLMs on AWS using Amazon SageMaker AI: Unified Studio, JumpStart, Training, Pipelines, Model Registry and Endpoints.
- Live Course
- 5 modules
-
Enterprise AI and agent development on Google Cloud Vertex AI with Gemini models: Google Gen AI SDK, grounding with Vertex AI Search, context caching, safety controls and deployment.
- Live Course
- 5 modules
-
Developing modular multi-agent systems with Google's open-source Agent Development Kit (ADK): LlmAgent, Sequential/Parallel/Loop workflow agents, MCP tools, local evaluation and deployment.
- Live Course
- 5 modules
-
Learn hands-on how to design, deploy, and operate scalable, secure, and managed enterprise agent architectures using Google Cloud Vertex AI Agent Engine and Agent Development Kit (ADK).
- Live Course
- 4 modules
-
Learn how to evaluate, customize, deploy as secure endpoints, and optimize cost-performance for Gemini, partner models, and open-weight models on Vertex AI Model Garden.
- Live Course
- 4 modules
Get in touch
