Learn how AI agents work, how to configure different model providers, connect tools with MCP, and add callbacks and streaming to your agent harness.

Agent Harnesses and The Agent Loop

Understanding Agent Harnesses

Model Providers

Configuring Model Providers

MCP & The Tools Ecosystem

Callbacks and Streaming

Tools and Callbacks

An agent harness is best described as

What happens after a tool finishes executing

Why use a different model for evaluations

What security risk with agent tools

Why suppress the default output handler

Agent Fundamentals

Control your agent's behavior with hooks, extend it with plugins and skills, and improve reliability using Strands Steering.

Hooks

Plugins & Skills

Steering

Hooks and Plugins

Keep Agents on the Rails with Steering

Hooks are most analogous to

What the agent sees when a hook cancels a tool call

Progressive disclosure in agent skills

Steering logic outcomes

LLM-based vs deterministic steering

Agent Control

Master context engineering, conversation management, and persistent memory with session managers.

Context Engineering & Conversation Management

Context Engineering

Session Managers

Persisting Conversation History

Sliding window conversation manager behavior

Summarizing vs sliding window managers

How session managers are implemented

Memory & Context

Build sophisticated multi-agent systems using agents-as-tools, graph workflows, and agent swarms.

Agents as Tools

Graph Workflows

Agent Swarms

Multi-Agent Architectures

Specialist agent context in agents-as-tools

Agent C runs after A and B complete

What happens in a swarm handoff

Multi-Agent Systems

Evaluate your agents thoroughly and deploy them to the cloud with Amazon Bedrock AgentCore.

Evaluating Agents

Evaluation Techniques

Deploying Agents to the Cloud

Deploying to Production with AgentCore

What tool filtering helps prevent

Preventing repeated failing tool calls

Primary benefit of plugin architecture for skills

What context offloading does with large tool results

Why agents-as-tools cannot guarantee invocation order

Safety risk unique to agent swarms

Why single-turn evaluation is insufficient

What a trajectory evaluator checks

What stays the same when deploying to the cloud

Production

Learn to design, build, evaluate, and deploy AI agent systems using the open-source Strands Agents SDK. Starting from core concepts like the agent loop and tool integration through to multi-agent orchestration and cloud deployment with Amazon Bedrock AgentCore.

Building AI Agent Harnesses with Strands Agents

Evaluation Techniques

Building AI Agent Harnesses with Strands Agents

実践的なインタラクティブ演習