Hands-on Agentic AI for Leaders¶

Configure your AI platforms to match your professional standards and communication style. You'll set up user preferences, define output formats, configure security settings including privacy controls and MFA, and establish baseline behaviors. By the end of this lesson, your AI tools will consistently deliver responses aligned with your professional requirements—without needing to specify preferences in every conversation.

Objectives:

• Configure personalized work profiles on your AI tools using custom instructions that reflect your role, work context, and response preferences
• Implement privacy and security protections across all platforms by disabling model training data sharing and enabling multi-factor authentication
• Apply personalization strategically across platforms based on each platform's available features (custom instructions, user preferences, style settings, output formats)

Activate AI memory systems that automatically retain your professional context across conversations. You'll enable persistent memory, populate your tools with key facts about your role and work, establish memory hygiene practices for reviewing and editing stored information, and set privacy boundaries. Your AI assistants will remember your context automatically—no more re-explaining who you are or what you're working on every conversation.

Objectives:

• Configure active memory systems on your AI platforms that retain key professional context across conversations without repeated explanations
• Implement memory settings across platforms by enabling persistent context between sessions
• Populate your AI memory with 2-3 critical professional facts about your role, work context, and communication preferences
• Apply memory hygiene practices by reviewing stored information and protecting sensitive data from AI memory

Transform your development workflow with an AI-powered code editor (Cursor or VS Code). Configure AI model integration (Claude, ChatGPT, Gemini, or other CLI models) that lets you generate code through natural language, get instant explanations for complex logic, and refactor with AI assistance. This workspace becomes essential infrastructure for building Skills, workflows, and Agents—enabling builders without deep coding backgrounds to create production-quality AI systems.

Objectives:

• Configure an AI-powered code editor (Cursor or VS Code) as your workspace for AI building block development
• Install and authenticate at least one AI model integration (Claude Code, ChatGPT Codex, Gemini CLI, or similar)
• Create and preview a Markdown file to verify your editor is properly configured

Objectives:

• Configure MCP connections to connect your AI assistant to external tools and data sources with proper authentication and permissions
• Set up working connectors for business applications (Gmail, Drive, Notion, Slack, HubSpot) relevant to your workflow
• Test each configured connector by executing prompts that successfully call external tools and retrieve data
• Distinguish between remote MCP servers (cloud-based) and local desktop extensions to select the right architecture for different use cases

Implement GitHub version control as your safety net for AI building block development. Create a repository to store Skills, Prompts, Agent configurations, and workflows with full change history. Master the create-commit-push workflow that professional developers use to track iterations, experiment without risk, collaborate across teams, and recover from mistakes—critical infrastructure for building production AI systems.

Objectives:

• Create a GitHub repository to store and version your AI building blocks (Skills, Prompts, Agents, configurations)
• Execute the create → commit → push workflow to save a building block to your repository
• Verify your repository syncs between local and cloud to protect your work

Install and configure a voice-to-text application (Wispr Flow recommended) to enable hands-free interaction with AI tools and code editors throughout the course. You'll practice voice dictation, learn best practices for speaking with AI, and set up your environment for maximum productivity. Voice-first collaboration dramatically accelerates your workflow—especially for complex prompts and multi-step instructions.

Objectives:

• Install and configure a voice-to-text application (Wispr Flow or alternative) with proper permissions and settings for AI platform integration
• Practice voice dictation techniques including formatting commands, punctuation control, and natural speech patterns to achieve 90%+ accuracy
• Demonstrate hands-free AI collaboration by completing a workflow entirely through voice—from initial prompt to refinement to final output
• Apply voice-first best practices for complex multi-step instructions, ensuring clarity and reducing the need for manual typing throughout the course

Map the agentic AI landscape — from deterministic automations to fully autonomous agents — and combine autonomy level with execution mode (augmented vs. automated) to classify any workflow on the AI Workflow Design Matrix.

Objectives:

• Distinguish between deterministic, guided, and autonomous workflows on the autonomy spectrum
• Explain the difference between augmented (human-in-the-loop) and automated (AI solo) execution modes
• Explain why agentic approaches outperform zero-shot prompting for multi-step tasks
• Classify example workflows on the AI Workflow Design Matrix (autonomy level × execution mode)

Learn the 10 building blocks of agentic AI — organized across Intelligence, Orchestration, and Integration layers — and practice identifying which blocks a workflow needs based on its classification on the design matrix.

Objectives:

• Describe the 10 AI building blocks and how they organize into three layers (Intelligence, Orchestration, Integration)
• Explain what each building block does and when to use it
• Distinguish between similar blocks (Skills vs MCP, Prompt vs Skill, Context vs Memory)
• Identify which building blocks a given workflow requires

Analyze your daily work for AI workflow opportunities using Step 1 (Analyze) of the Business-First AI Framework. Run a structured AI-facilitated audit to surface candidates, classify each on the AI Workflow Design Matrix by autonomy level and execution mode, and select 2-3 candidates that span the matrix as your building projects for the rest of the course.

Objectives:

• Explain Step 1 (Analyze) of the Business-First AI Framework as a structured method for identifying AI workflow opportunities in your work
• Identify workflow candidates from your actual work by running a structured AI-facilitated audit using the Analyze method
• Classify each candidate on the AI Workflow Design Matrix by autonomy level and execution mode with reasoning
• Select 2-3 workflow candidates that span the design matrix — prioritized by impact, frequency, and feasibility — as your building projects for the course

Turn implicit workflows into structured specifications you can build from. You'll deconstruct two real-world workflows—one Augmented + Guided, one Automated + Deterministic—using the 5-question framework to surface discrete steps, decision points, data flows, context needs, and failure modes hiding inside each process. Then apply the same framework to your own workflow, producing a Workflow Definition that becomes your blueprint for Step 3 (Build).

Objectives:

• Apply the 5-question deconstruction framework to break down a workflow into detailed steps, surfacing the decision points, data flows, context needs, and failure modes that are often invisible in day-to-day execution
• Generate a structured Workflow Definition that captures every step with sufficient detail to serve as a build-ready specification
• Compare the deconstruction output of two contrasting workflow types (e.g., Augmented + Guided vs. Automated + Deterministic) to identify how step structure, decision complexity, and context needs differ across the AI Workflow Design Matrix

Understand why workflow inventories are the foundation of AI operational control — and how the three-level hierarchy (Business Process → Workflow → Building Block) organizes everything you build. You'll see how Process Guides, SOPs, and Building Blocks connect into a navigable system, and examine the hierarchy in action through a real course development registry. This conceptual foundation prepares you for the next project, where you'll build your own AI Operations Registry.

Objectives:

• Explain why workflow inventories matter for AI operational control — leaders can't manage AI-first processes they haven't mapped
• Distinguish between business processes, workflows, and building blocks as three levels of an operations hierarchy — strategic, operational, and tactical
• Identify the artifact types that document each level — Process Guides (when/why), SOPs (how to execute), and Building Blocks (prompts, skills, agents, context)
• Describe how a registry links these levels into a living system that enables impact analysis, reuse tracking, and gap identification

Dissect what Skills are—reusable instruction sets that teach Claude how to execute your workflows consistently. Analyze their anatomy: metadata, instructions, and resources. Then distinguish when to use Skills versus Prompts, Projects, MCP, or Agents.

Objectives:

• Analyze the anatomy of an Agent Skill including metadata, instructions, and resource structure
• Distinguish Skills from Prompts, Projects, MCP, and Subagents to select the right building block
• Identify when to use each agentic building block based on the workflow requirements

Transform your workflow knowledge into a production-ready Agent Skill from scratch. You'll structure your skill with proper metadata and instructions, test it in a live conversation, save it to your repository with version control, and export it to your local machine—making it available across Claude.ai, Claude Code, and Cowork.

Objectives:

• Build a production-ready Agent Skill that packages your expertise into reusable instructions for Claude
• Structure your skill with proper metadata, instructions, and supporting resources
• Save skills to your repository for version control and register them in your AI Assets database
• Export skills to your local machine for use across Claude platforms (Code, Cowork)

Package and ship your skills as installable resources. One command gives others access to your expertise—team members, clients, or the broader AI community. Your workflow knowledge becomes reusable, distributable, scalable.

Objectives:

• Create a properly structured marketplace.json file with plugin metadata for at least one skill
• Publish a GitHub repository containing their skill marketplace that others can add via /plugin marketplace add
• Configure distribution settings for three deployment scenarios: private team use, organizational access, and public community sharing
• Demonstrate the complete installation workflow by having another team member successfully install their published plugin

Systematically measure and improve your Agent Skills using structured evals. You'll write test cases with expected outcomes, benchmark pass rates and token usage, compare skill versions with blind evaluation, and optimize skill descriptions for triggering accuracy — learning to refine skills with data, not intuition.

Objectives:

• Evaluate skill quality by writing test cases with expected outcomes, running evals in isolated contexts, and benchmarking pass rates and token usage — completing at least one full test-measure cycle that produces quantitative performance data.
• Compare skill versions using blind A/B evaluation to determine whether changes actually improved performance, making refinement decisions based on evidence rather than intuition.
• Optimize skill descriptions for triggering accuracy by analyzing activation patterns against test prompts, reducing false positives (overly broad triggers) and false negatives (missed activations) to improve when the skill fires.

A self-study reference resource for understanding AI workflow execution patterns. Covers the mental models that determine which AI workflow pattern fits your use case — from deterministic prompt chains to fully autonomous agents — and the Pattern Selection Framework for matching workflow characteristics to the right architecture.

Objectives:

• Distinguish between seven workflow architecture patterns and five agent capability patterns by categorizing them as execution structure vs. behavioral capabilities
• Analyze workflow characteristics (predictability, complexity, step count) to select appropriate patterns using the Pattern Selection Framework.

Design your workflow's AI implementation by mapping each step to specific building blocks, autonomy levels, and orchestration patterns, producing a Building Block Spec ready for the Build phase.

Objectives:

• Design a Building Block Spec for at least one workflow by translating each step in their Workflow Definition into building block assignments, orchestration decisions, and human involvement gates, producing a complete implementation blueprint ready for the Build phase.
• Construct an orchestration design for their workflow by selecting the mechanism (Prompt, Skill-Powered Prompt, or Agent) and execution mode (Augmented or Automated) that best fits the complexity, variability, and human judgment requirements of each step.
• Develop step-level building block assignments for every step in their Workflow Definition by mapping each step's decision complexity, context needs, and tool dependencies to specific autonomy levels and AI building blocks.

Construct and run deterministic workflows using the skill-powered prompt orchestration pattern. You'll follow the Build framework to create workflows with fixed steps, clear input/output definitions, and consistent outputs — then test them on real scenarios and validate that they produce reliable results without human intervention.

Objectives:

• Build deterministic workflows using skills or skill-powered prompts that execute fixed steps with clear input/output definitions, producing consistent outputs from the same inputs across multiple runs.
• Configure an automated trigger (schedule or event) for a workflow, demonstrating the automated execution mode where no human is in the loop during execution.
• Test the workflows on real scenarios and verify output consistency across at least two runs, identifying and fixing any steps that produce variable results.

Construct and run guided workflows where AI operates within defined boundaries while you provide judgment at key checkpoints. You'll build skill-powered prompts with explicit review points, structure handoffs between AI execution and human refinement, and create feedback loops — producing workflows in a different Design Matrix cell from your deterministic builds.

Objectives:

• Build guided workflows using skill-powered prompts with explicit checkpoints where human judgment steers AI execution, producing working workflows in the guided column of the Design Matrix.
• Design handoff points that structure the flow between AI-generated drafts and human review, with clear decision criteria for when to accept, refine, or redirect outputs.
• Compare the guided workflows to the deterministic workflows built earlier in the session, articulating at least two differences in how autonomy level affects orchestration design, human involvement, and output variability.

Objectives:

• Configure a ChatGPT agent by defining a clear goal, selecting appropriate tools, and providing necessary context to create an autonomous agent that executes multi-step tasks independently
• Formulate effective agent goals that are specific enough to guide execution while allowing the agent flexibility to plan and adapt its approach dynamically
• Observe and analyze agent execution by examining the agent's reasoning process, tool selection, and decision-making patterns to understand how it autonomously achieves the goal
• Troubleshoot agent behavior by identifying common failure modes (over-planning, tool misuse, scope creep) and refining the goal and context to improve reliability and performance

Objectives:

• Configure ChatGPT Atlas for browser-based automation
• ◦ Complete initial Atlas setup and authentication
• ◦ Configure the Atlas browser extension with appropriate permissions
• ◦ Adjust security and privacy settings for workflow execution
• Execute autonomous browser workflows using Atlas agent mode
• ◦ Navigate websites using Atlas autonomous mode
• ◦ Delegate multi-step browser tasks to Atlas agents
• ◦ Monitor agent execution and intervene when necessary
• Apply Canvas browser queries to extract website information
• ◦ Ask contextual questions about webpage content
• ◦ Extract structured data from websites using Canvas queries
• ◦ Distinguish when to use Canvas queries vs. autonomous agent mode

Analyze the core components every AI agent shares—LLM brain, tools, memory, instructions, and knowledge—then distinguish agents from structured workflows on the agentic systems spectrum to determine when autonomous agents are the right solution for a business problem.

Objectives:

• Distinguish between agents and workflows by comparing their autonomy levels, decision-making capabilities, and appropriate use cases on the agentic systems spectrum
• Identify the core components of an agent (LLM brain, tools, memory, instructions, knowledge) and explain how these components work together regardless of platform
• Evaluate business scenarios to determine when autonomous agents are appropriate versus structured workflows, based on task predictability, complexity, and need for dynamic planning
• Select optimal use cases for agentic implementation by matching task characteristics (open-ended problems, multi-step complexity, tool requirements) to agent capabilities

Distinguish the five agentic capability patterns—Reflection, Tool Use, Planning, Multi-agent Collaboration, and Memory—then analyze how they work together in real scenarios and differentiate guardrails from human-in-the-loop controls for managing agent behavior.

Objectives:

• Distinguish the five agentic capability patterns (Reflection, Tool Use, Planning, Multi-agent Collaboration, Memory) by mapping each to its specific role in the agent execution cycle
• Analyze how multiple capability patterns combine in the customer exchange scenario to identify which patterns drive each step of agent behavior
• Differentiate guardrails from human-in-the-loop controls by identifying when each is appropriate for managing agent risk

Analyze four orchestration patterns for multi-agent systems—supervisor, swarm, hierarchical, and debate—then compare their strengths and tradeoffs to select the right pattern for different workflow scenarios.

Objectives:

• Analyze orchestration patterns for multi-agent systems: supervisor, swarm, hierarchical, and debate
• Compare pattern strengths and weaknesses for different task types and complexity levels
• Select the appropriate orchestration pattern for a given multi-agent workflow scenario

Apply the Business-First AI Framework to deconstruct your own workflow into discrete steps, map them to AI building blocks, and create a complete platform-agnostic agent design specification ready to implement on Claude Code, ChatGPT, Gemini, or any platform.

Objectives:

• Apply a systematic design process for agent systems by defining clear goals, processes, agent roles, instructions, and required tools
• Generate clear, unambiguous instructions for LLM agents using meta-prompting techniques and reasoning models to ensure reliable execution
• Design single-agent systems by specifying agent roles, responsibilities, and tool requirements for focused automation tasks
• Architect multi-agent systems by identifying and naming specialized agent roles that work together to accomplish complex goals
• Distinguish between single-agent and multi-agent approaches based on task complexity and the need for specialized expertise across different workflow components

Understand what Claude Code subagents are, master the critical distinction between Skills and Subagents, and learn when to use each approach. Explore the mental model of building a specialized team of AI agents, develop a decision framework for choosing between Skills and Subagents, and identify appropriate use cases for subagent-based solutions in business contexts.

Objectives:

• Define Claude Code subagents and explain their role in autonomous workflows
• Distinguish between Claude Skills (training manuals with shared context) and Subagents (specialized employees with isolated context) across key dimensions
• Apply a decision framework to determine when to use Skills, Subagents, or both for specific workflow requirements
• Analyze the "dream team" mental model to understand subagent specialization and delegation patterns
• Evaluate business scenarios to identify appropriate use cases for subagent-based autonomous workflows

Transform your agent system design into a working implementation. Configure MCP tools in Claude Code, create your subagent with appropriate tool assignments and instructions, and test it with real tasks. Learn to debug, iterate, and refine subagent behavior through hands-on practice. Students complete this lesson with a fully functional subagent ready for real-world use.

Objectives:

• Configure MCP tools in Claude Code to prepare tool access for subagent implementation
• Create a subagent by translating design specifications into Claude Code configuration (description, tools, model, system prompt)
• Evaluate subagent performance by testing with real tasks and observing delegation behavior
• Refine subagent configuration based on testing results to improve reliability and output quality

Master subagent execution from single-agent workflows to complex multi-agent orchestration. Learn three activation methods (explicit invocation, automatic delegation, chaining), understand when multiple agents outperform single-agent solutions, and apply orchestration patterns including sequential, parallel, and hybrid execution with hooks for reliable autonomous workflow behavior.

Objectives:

• Execute Claude Code subagents using three activation methods (explicit invocation, automatic delegation, chaining) to accomplish single-task workflows
• Evaluate when to use multiple specialized agents versus a single agent based on workflow complexity and context isolation needs
• Design a multi-agent workflow using sequential, parallel, or hybrid orchestration patterns for a complex business use case
• Implement hooks (guardrails, logging, human-in-the-loop) to ensure reliable multi-agent system behavior

Build complete multi-agent systems from your workflow specifications using Claude Code's AI-assisted configuration. You'll transform your Module 10 design into a working system—single or multi-agent—by prompting Claude Code to generate all necessary subagent configurations. Learn to review, refine, and validate auto-generated systems, ensuring proper orchestration and tool delegation. Master the AI-assisted workflow that makes complex agent systems buildable in minutes.

Objectives:

• Build complete agent systems by prompting Claude Code to auto-generate configurations from workflow specifications
• Evaluate generated subagent configurations against design intent to ensure proper tool assignments and orchestration patterns
• Refine multi-agent systems by adjusting delegation logic, tool restrictions, and coordination patterns based on testing results
• Apply judgment to decide system architecture (single vs. multi-agent) and when to use AI-assisted generation versus manual configuration

Learn to automate Claude Code subagent execution using native operating system schedulers. Configure Windows Task Scheduler and macOS cron jobs to run subagents on recurring schedules, implement error handling and notifications, and understand when scheduled execution is appropriate versus on-demand triggers. Students will schedule one of their existing Claude Code subagents to run automatically.

Objectives:

• Configure OS-level scheduling (Windows Task Scheduler or macOS cron) to execute Claude Code subagents automatically
• Implement error handling and notification mechanisms for scheduled subagent runs
• Evaluate which workflows benefit from scheduled execution versus on-demand triggers
• Test and verify scheduled subagent execution with proper logging and output capture

Master Claude Code's built-in tracing capabilities to gain visibility into subagent execution. Learn to read execution traces, understand multi-agent delegation patterns, diagnose failures, and debug common issues. Students will add tracing to their multi-agent systems and use trace data to troubleshoot and optimize subagent behavior.

Objectives:

• Analyze Claude Code execution traces to understand subagent delegation patterns and tool usage
• Diagnose subagent execution failures using trace data and error logs
• Evaluate subagent performance and identify optimization opportunities through tracing
• Implement tracing best practices for debugging multi-agent subagent workflows