The roadmap

This course is built around four transitions. Each one is load-bearing. Each one builds on the one before. And each one changes what you physically do with AI tools.

Transition 1: Making the organization legible. Can the company describe its own work to a machine? This is where Cornwall Market starts. Sarah will walk through her store, dictating categorization rules into her phone. She will turn tribal knowledge into structured documents called skills. The output is not a demo — it is a spreadsheet of vendor mappings and a document that explains what account codes mean and why.

Transition 2: Trusting your own data. Can people act on what the system tells them? Sarah will connect Cornwall Market’s actual data — invoices, QuickBooks records, supplier lists — to Claude through MCP, the Model Context Protocol. She will test the system against real invoices and discover that it catches errors she was missing. In a parallel deployment at Fountain Creek (a multi-location food service company), this same pattern hit 94.5% accuracy — and the machine caught human errors the humans had missed. Trust inverted.

Transition 3: The system starts acting. Does the system surface insights before being asked? Cornwall Market’s invoice processing moves from manual to semi-automated. Instead of Sarah opening each PDF, reading line items, and typing data into accounting software — 10 to 15 minutes per invoice, errors common — the system processes incoming invoices, flags exceptions, and presents categorized results for review. Delegation tools like Dispatch and cloud routines make this always-on.

Transition 4: The system changes how the org works. Does the system learn from its own operation? When Sarah corrects a categorization, that correction feeds back into the system’s knowledge. When a new supplier appears, the system proposes an account mapping based on patterns it has learned. The organization and the system co-evolve.

Here is the concrete map of what you will do, chapter by chapter:

The transitions are sequential. You cannot skip them. But you can start from wherever you are.

There is a second thread running through these chapters that is just as important as the transitions: the distillation pipeline. Every conversation you have with Claude is a potential source of organizational knowledge. The course follows a progression from the most informal capture to the most formal:

Each step increases the formality, the shareability, and the automation. A conversation helps one person in one session. Project instructions help every chat in one workspace. A SKILL.md helps every project that installs it. An MCP server makes the knowledge available across every Claude surface. A habitat closes the loop — corrections flow back in and the system improves itself.

The key insight: don’t just ask Claude to do things. Ask it to document how it did them so it can do them again. Every discovery gets documented back into the system. That is how a conversation becomes an agent.

The maturity framework

Before you move forward, figure out where you stand. The maturity levels below describe how an organization relates to AI systems — not how many AI tools it has purchased, but how deeply AI is woven into how work actually gets done.

Cornwall Market is solidly L0. The categorization rules live in Sarah’s head. Supplier relationships are personal — she knows that Chen’s Produce always rounds up on invoices by a dollar or two, and it is fine because they have worked together for twelve years. Exception handling is oral tradition: “If the broad-line distributor ships a substitution, call them before you pay the invoice.” “If a bakery supplies order is over $800, double-check because they sometimes double-ship flour.” None of this is written down. None of it is shareable. If Sarah wins the lottery tomorrow, Cornwall Market has a real problem.

Three surfaces, one connective tissue

This course uses Claude’s ecosystem as the worked example. Not because it is the only option, but because it is the most complete ecosystem available right now — and because demonstrating each transition requires real tools that actually work together.

Claude’s ecosystem has three surfaces. Think of them as three different ways to work with the same underlying intelligence:

The surfaces are a progression. You start with Claude.ai for conversations, capture, and artifacts. You move to Cowork when you need to connect data and build workflows. You move to Claude Code when you are ready to build systems that run in production. Each chapter will tell you which surface to use and why.

Around these three sits a wider family you will meet along the way: Claude in Chrome (Claude operating your browser from a side panel), Claude Code on the web (the same agent running on cloud infrastructure at claude.ai/code, steerable from any device), IDE extensions for VS Code and JetBrains, @Claude in Slack, GitHub Actions for CI, and the Agent SDK for teams building Claude into their own products. They are not more things to learn — they are more doors into the same system. That is the point of this course’s architecture: knowledge and connections you set up once become available through every door.

Connecting all three surfaces is MCP — the Model Context Protocol. MCP is a standard that lets any tool, service, or data source expose itself to Claude through a single interface. It is the connective tissue that makes your QuickBooks data available in Claude.ai, your invoice PDFs accessible in Cowork, and your production workflows operational in Claude Code. You will meet it gently in Chapter 05 and wire it up for real in Chapter 07.

The framework — the four transitions, the maturity levels — applies regardless of which AI vendor you use. The implementation details are specific to Claude because that is where the tools are most mature. If you are using a different ecosystem, translate the concepts. The organizational work is identical.

Check your understanding

Answer in your own words — write it down before opening the key. Your tutor grades against the criteria and generates fresh variants on retries.

Q1. A 12-person law firm: two partners use ChatGPT personally with their own prompts, nothing is shared, and the knowledge of which precedents matter for which client types lives in the senior partner’s head. Where is this firm on the maturity ladder, and what — concretely — does their next transition require?

Q2. Pick the right surface and defend it: (a) analyzing a folder of 200 case files for inconsistencies, (b) a quick question while waiting for coffee, (c) a job that must run every night at 2am. One sentence of justification each.

Next: Install the App