This lesson introduces the fundamentals of text generation with Claude, showing how to create written content such as blog posts, articles, product descriptions, and professional documents. It emphasizes an iterative and conversational approach to refining drafts, moving from outlines to polished paragraphs, and highlights how to guide tone, structure, and audience focus. Learners explore how Claude supports various writing styles and use cases to enhance creativity and productivity in content creation.

This lesson surveys non-text content types you can create with Claude, including code artifacts, utilities, HTML and SVG visuals, Mermaid diagrams, and interactive dashboards or components. It shows how to iterate on features, preview and version outputs, and publish artifacts for safe sharing and reuse.

This lesson explores content refinement and the different input types supported by Claude. It demonstrates how to remix and improve existing artifacts, incorporate feedback, and work with multimodal inputs such as text, images, and documents. Learners discover how Claude processes various file formats and how these can be used to enhance projects, analyze data, and refine outputs for more effective results.

This lesson outlines why and how to upgrade to Claude Pro. It explains the benefits of higher usage limits, larger context capacity, support for richer file workflows, and access to Projects for building knowledge bases. You see the upgrade flow, then a demonstration of working with bigger datasets and interactive dashboards, plus notes on team options for collaboration.

This lesson explores how Claude Pro allows you to create and apply custom writing styles. It demonstrates uploading examples of your own work to train Claude to match your tone, as well as using preset styles or mimicking well-known voices. Learners see how to switch between styles, refine outputs, and combine influences to develop a distinct, personalized writing voice.

This lesson demonstrates how to connect Google Drive to Claude Pro for use in chats and projects. It walks through the setup process, permission prompts, and how to analyze or refine writing directly from Google Docs. Learners see how Claude can summarize, review, and edit documents, and learn the current limitations, including support for Docs only and syncing updates automatically.

This lesson shows how to create a custom Claude Project, add clear project instructions, and organize project knowledge. It demonstrates building an interactive practice test artifact with timed questions, navigation, and scoring, then iterating to refine behavior. Learners see how project chats, files, and prompting guides work together to deliver reusable workflows.

This lesson continues refining the custom Claude Project by enhancing an interactive practice test. It demonstrates iterative prompting to fix logic issues, add review and scoring features, and capture improvements in a knowledge article. Learners see how to document and update project knowledge so enhancements persist across future sessions, reinforcing best practices for building and maintaining reusable Claude projects.

This lesson shows how to expand a custom Claude Project knowledge base and make improvements persist. You learn to add long outputs in parts, attach artifacts to the project, manage file size and capacity limits, upload PDFs and docs, and use that knowledge to automatically generate comprehensive practice tests with timers, then iterate for accuracy and quality.

A live, unscripted build (first of two parts). Watch Mike analyze a prior Claude Code session that

produced low-quality Canva slides, then set up Claude Cowork as the bridge between chat and code, build a

Velza brand kit inside Canva from scratch, and hit Canva MCP's 90-character description wall plus 12

consecutive generation timeouts.

You'll learn the three-layer compression pipeline that degrades AI design output — the 90-character wall,

lossy style descriptions, and MCP server timeouts — and why "brand kit first" is the #1 lesson. The

session ends with the HTML fallback and the realization that one-shotting a 17-slide presentation is

doomed.

Tools used: Claude Desktop (Opus 4.6), Claude Cowork, Claude Code, Canva Pro, the Canva MCP connector.

After hitting Canva's walls (90-character description limit and 12 consecutive timeouts), Mike pivots from one-shot prompting to a chain-of-thought multi-agent workflow using his Agent Team Requirements Document (ATRD) framework.

Watch Claude Cowork produce the requirements doc, then Claude Code spawn a five-phase agent team — Writer, Builder, Editor, Assembler, Exporter, with batched 5-slide generation, four-tier retry logic, and a content

manifest JSON that lets any agent pick up where another left off. The session covers the one-liner that

launches the entire pipeline, batch-one success with four style candidates to choose from,

style-vs-substance taste calls, and the token-cost reasoning that decides when to let the agent push

through versus when to fix something manually in Canva.

A live demonstration of the "human at the helm" principle in a real production workflow.

Tools used: Claude Cowork, Claude Code with the --dangerously-skip-permissions flag (in a sandboxed

Cowork folder), Canva Pro, the Canva MCP connector, and requirements markdown skill.

Here's a live build where I leverage Claude Code and Cowork to connect to the Canva Connector with Claude via the Canva MCP server. I leverage an Agent Team Requirements Document (ATRD) to work through the creation of an executive presentation that is 17 slides long. Along the way, we encounter a lot of limits that Canva throws at us, and figure out a way to perform a chain-of-thought back-and-forth prompting technique to get the slides built within the constraints and limits that the Canva API and MCP Server have in place. This allows for a separation of concerns between two AI tools - Claude and Canva, letting each do what they do best. As we starting hitting limits and the presentation quality falls off the cliff, I figure out that Canva is communicating what it can and cannot do, back to Claude, sending its limitations. This is a great example of Human-at-the Helm concepts as to where human judgement kicks in to get things to work and to get these two AI systems to give me what I want, ultimately.