Agentic AI with Claude Training - Canada

Introduction to Agentic AI

1. What Agentic AI means
2. Generative AI vs. Agentic AI
3. Chatbots vs. assistants vs. workflows vs. agents
4. How agents move from answering to taking action
5. Understanding goals, actions, observations, and feedback
6. Common characteristics of AI agents
7. Where agentic AI provides real business value
8. When an AI agent is unnecessary

Understanding the Agentic Loop

1. Receiving a goal or task
2. Understanding the current context
3. Planning the next action
4. Selecting an appropriate tool
5. Executing an action
6. Observing the result
7. Evaluating progress
8. Replanning when necessary
9. Completing or escalating the task
10. Understanding stopping conditions

Introduction to Claude

1. Understanding Claude and its role in generative AI
2. Working effectively with Claude
3. Claude for reasoning and business tasks
4. Claude for document and knowledge workflows
5. Claude for structured information
6. Claude for coding and technical tasks
7. Claude as the reasoning layer of an AI agent
8. Understanding model capabilities and limitations

Foundations of Effective Claude Prompting

1. Writing clear instructions for Claude
2. Defining roles and objectives
3. Providing business context
4. Defining constraints and boundaries
5. Specifying output requirements
6. Supplying examples
7. Working with long context
8. Iteratively improving Claude responses
9. Creating reusable prompt structures

Designing System Instructions for Agents

1. User prompts vs. system-level instructions
2. Defining an agent's role
3. Establishing agent responsibilities
4. Defining what the agent can and cannot do
5. Establishing decision rules
6. Controlling tone and behavior
7. Defining escalation conditions
8. Preventing agents from acting outside their scope
9. Building reusable agent instruction templates

Structured Outputs

1. Why structured outputs matter in agentic systems
2. Turning natural-language responses into structured information
3. Designing predictable output schemas
4. Extracting fields from unstructured information
5. Generating JSON-style structured responses
6. Validating AI-generated outputs
7. Passing structured outputs between workflow stages
8. Using structured information for automation

From Prompting to AI Workflows

1. Understanding deterministic workflows
2. Sequential AI workflows
3. Conditional workflows
4. Routing tasks based on context
5. Combining multiple AI steps
6. Passing outputs between tasks
7. Building reusable workflow components
8. Understanding where agents should replace fixed workflows

Understanding AI Tools

1. What tools mean in agentic AI
2. Giving Claude capabilities beyond text generation
3. Search tools
4. Database tools
5. File and document tools
6. Calculation tools
7. Business application tools
8. Internal APIs
9. External APIs
10. Understanding tool descriptions and parameters

Tool Use with Claude

1. How Claude decides when a tool is required
2. Defining tools for an AI agent
3. Providing tool descriptions
4. Defining tool inputs
5. Receiving tool results
6. Using tool results in subsequent reasoning
7. Handling multiple tool calls
8. Preventing unnecessary tool usage
9. Building reliable tool-use loops

Claude API Fundamentals

1. Understanding APIs
2. Introduction to the Claude API
3. Sending messages programmatically
4. Providing system instructions
5. Managing user and assistant messages
6. Working with model responses
7. Understanding request and response structures
8. Handling errors
9. Managing API credentials securely
10. Basic cost and usage considerations

Building Your First Claude Agent

1. Defining an agent objective
2. Creating system instructions
3. Connecting available tools
4. Designing the agent loop
5. Receiving user requests
6. Selecting actions
7. Calling tools
8. Processing tool results
9. Producing final answers
10. Testing the completed agent

Planning and Task Decomposition

1. Breaking complex goals into smaller tasks
2. Creating task plans
3. Sequential task execution
4. Dynamic replanning
5. Handling incomplete information
6. Asking users for missing information
7. Managing dependencies between tasks
8. Avoiding unnecessary over-planning
9. Determining when a task is complete

Reasoning and Decision-Making

1. Using Claude for decision support
2. Comparing available actions
3. Applying business rules
4. Handling uncertain information
5. Prioritizing tasks
6. Evaluating intermediate results
7. Recovering from unsuccessful actions
8. Knowing when human input is required

Retrieval-Augmented Agent Workflows

1. Understanding Retrieval-Augmented Generation
2. Connecting agents with business knowledge
3. Retrieving relevant information before answering
4. Working with documents
5. Searching internal knowledge
6. Using retrieved information as context
7. Grounding agent responses
8. Reducing unsupported answers
9. Building knowledge assistants

Working with Documents

1. Processing business documents using Claude
2. Extracting structured information
3. Summarizing documents
4. Comparing documents
5. Identifying important clauses or sections
6. Classifying incoming documents
7. Creating document-processing workflows
8. Combining document understanding with agent actions

Agent Memory

1. What memory means in AI agents
2. Short-term conversational context
3. Long-term application memory
4. User preferences
5. Task history
6. Storing useful information
7. Retrieving relevant memories
8. Deciding what should not be remembered
9. Avoiding unnecessary or sensitive memory storage

Introduction to Model Context Protocol (MCP)

1. What MCP is
2. Why standardized AI-tool connectivity matters
3. Understanding MCP clients and servers
4. Connecting Claude with external tools and resources
5. Providing agents with controlled access to business systems
6. Exploring MCP-based workflows
7. Security considerations when exposing tools through MCP

Building Claude + MCP Workflows

1. Connecting an MCP-enabled tool
2. Discovering available capabilities
3. Allowing Claude to select appropriate tools
4. Retrieving external information
5. Performing actions through connected systems
6. Combining several tools within one workflow
7. Controlling agent permissions
8. Testing MCP-enabled agents

Human-in-the-Loop Agent Design

1. Why autonomous does not mean unsupervised
2. Identifying high-risk actions
3. Creating approval checkpoints
4. Requesting confirmation before actions
5. Escalating uncertain decisions
6. Separating recommendations from execution
7. Designing reviewer workflows
8. Keeping humans responsible for critical decisions

Guardrails and Agent Boundaries

1. Defining permitted actions
2. Defining prohibited actions
3. Data-access restrictions
4. Limiting tool permissions
5. Preventing unauthorized actions
6. Validating tool inputs
7. Limiting execution scope
8. Preventing runaway workflows
9. Designing safe failure behavior

Handling Agent Failures

1. Why AI agents fail
2. Incorrect tool selection
3. Missing context
4. Invalid tool parameters
5. Hallucinated information
6. Repeated agent loops
7. Incomplete task execution
8. Conflicting instructions
9. Tool failures
10. Designing fallback strategies

Debugging Agentic Workflows

1. Tracing agent actions
2. Reviewing tool calls
3. Examining intermediate outputs
4. Finding weak system instructions
5. Identifying workflow bottlenecks
6. Detecting repetitive behavior
7. Testing edge cases
8. Improving reliability through iteration

Evaluating AI Agents

1. Why chatbot-style testing is insufficient
2. Defining successful agent behavior
3. Task-completion rate
4. Output accuracy
5. Tool-selection accuracy
6. Reliability
7. Latency
8. Cost
9. Safety
10. Human intervention rate
11. Developing repeatable evaluation scenarios

Single-Agent vs. Multi-Agent Systems

1. Understanding single-agent architectures
2. Understanding multi-agent systems
3. Specialized agent roles
4. Coordinator and worker patterns
5. Passing tasks between agents
6. Benefits of specialization
7. Coordination challenges
8. Increased cost and complexity
9. Knowing when one capable agent is better than multiple agents

Agentic AI for Business Research

1. Automated research workflows
2. Gathering information from multiple sources
3. Summarizing research
4. Comparing alternatives
5. Identifying trends
6. Generating research briefs
7. Highlighting missing evidence
8. Combining AI research with human verification

Agentic AI for Marketing and Content

1. Content research agents
2. Content planning workflows
3. Draft generation
4. Brand-guideline checking
5. Content repurposing
6. SEO research assistance
7. Campaign idea generation
8. Human approval before publishing

Agentic AI for Customer Support

1. Customer-query classification
2. Knowledge retrieval
3. Suggested responses
4. Ticket routing
5. Escalation workflows
6. Customer-history context
7. Human handoff
8. Controlling customer-facing agent actions

Agentic AI for Business Operations

1. Processing routine requests
2. Generating reports
3. Updating structured records
4. Creating task lists
5. Document processing
6. Knowledge retrieval
7. Internal workflow assistance
8. Automating repetitive administrative processes

Agentic AI for Data and Decision Support

1. Turning questions into analytical tasks
2. Collecting relevant information
3. Comparing options
4. Identifying trends
5. Creating management summaries
6. Generating decision briefs
7. Highlighting assumptions
8. Supporting rather than replacing human decision-makers

Security and Privacy for AI Agents

1. Understanding agent security risks
2. Protecting API keys and credentials
3. Handling confidential business data
4. Managing system permissions
5. Limiting access to external tools
6. Protecting customer information
7. Avoiding unnecessary data exposure
8. Applying least-privilege principles

Responsible Agentic AI

1. Human accountability
2. Transparency
3. Bias and fairness
4. Accuracy and hallucination risk
5. Privacy
6. Intellectual property considerations
7. Responsible automation
8. High-risk business decisions
9. Maintaining appropriate human oversight

Designing a Production-Ready Agent

1. Defining the business objective
2. Identifying users
3. Selecting required information
4. Selecting required tools
5. Designing agent instructions
6. Defining permissions
7. Establishing human approval points
8. Planning failure handling
9. Establishing evaluation criteria
10. Measuring business value

Practical Project: Build a Claude-Powered AI Agent

Participants design and build an agentic AI solution for a real-world scenario such as:

1. Business research agent
2. Customer-support assistant
3. Document-processing agent
4. Marketing research and content agent
5. Internal knowledge assistant
6. Business reporting agent
7. Operations assistant

The project includes:

1. Business problem definition
2. Agent goal
3. System instructions
4. Tool selection
5. Workflow design
6. Structured outputs
7. Retrieval or external knowledge
8. Human approval points
9. Guardrails
10. Error handling
11. Testing
12. Evaluation
13. Final demonstration

From Prototype to Business Implementation

1. Moving from experiments to usable workflows
2. Starting with narrow, high-value problems
3. Measuring time saved
4. Measuring quality improvements
5. Estimating operational costs
6. Identifying adoption barriers
7. Training employees to supervise agents
8. Improving agents using real-world feedback
9. Gradually expanding autonomy
10. Building an agentic AI adoption roadmap