The Tools Are Ahead of Us. Training Closes the Gap.
Why a training curriculum matters more than the next tool you buy, and the four-level program Dew Wealth deploys to build real AI fluency across a team.
The Capability Overhang
The AI community calls the gap between what today's models can already do and what people actually use them for the capability overhang. The frontier isn't waiting on better technology; it's waiting on us.
Tooling alone doesn't close it: license every tool on the market, and output stays flat if your team works the way it did last year. Fluency does, built deliberately, one person and one level at a time. Below is the curriculum Dew Wealth runs internally: not a tour of features, but a path from basic chat use to building skills, automations, and designs, with a capstone to defend at every level.
Four Principles Run Through Every Course
AI Drafts, You Think
Every output that leaves your hands gets read and edited by a human who understands it. If a draft isn't worth your time to read and fix, it isn't worth a colleague's or a client's time either. That is the anti-slop standard, and it is non-negotiable.
Don't Outsource the Learning Itself
There is a real risk (cognitive offloading, and the cognitive debt that follows) in leaning on AI while you are still learning something new. Use AI to go faster on things you already understand; be careful using it as a crutch for things you don't.
The Best Tutor for Claude Is Claude
Everyone in the firm runs their own professional development coach and coaching projects: a place to ask questions, get unstuck, and pressure-test thinking. You build these early and use them throughout.
Comprehension Is the Hard Part
Producing something is easy now. The real work in every course is testing and validating that you, and the model, actually got it right. Each level ends in a capstone you have to build and then defend.
Two disciplines show up repeatedly: context discipline(enable only the tools, skills, and plugins a task needs) andrisk awareness (what you connect through MCP and what you put into a prompt both carry data exposure).
Four Levels, One Path
Each level ends in a capstone you have to build and then defend. Expand any course for its description.
Basic AI Chat Usage
Foundational fluency. Before anyone builds anything, they learn where AI helps, how to set up their workspace, how to prompt, and how to stay efficient. Judgment first, then output.
When to Use AI (and When Not to)
The judgment call at the center of everything else: where AI genuinely helps and where it quietly hurts. Covers cognitive offloading and cognitive debt, and why leaning on AI while onboarding or learning a new skill can stunt the very ability you are trying to build. Introduces the firm's coaching model and frames AI risk awareness and the basics of MCP connectors. Homework is to stand up your coach projects and show proof you are already using them.
Basic Harness Setup
Vocabulary and setup before you touch a prompt. Defines what a "harness" is and the core terms used for the rest of the program. Claude Chat versus Claude Cowork: what each one is for and when to reach for which. Walks through configuring your system instructions so the model starts every task with the context it needs instead of guessing.
Basic Prompt Engineering
How to get a usable output on the first or second pass instead of the tenth. How to use the prompt engineer, structure a request, and give the model what it needs up front. The payoff is concrete: better output quality and far fewer wasted tokens lost to vague prompts and unnecessary back-and-forth.
Claude Chat Projects
How to organize your work in Projects. When to spin up a new project versus keeping a master project you return to; what belongs in reference files versus system instruction detail; and the limits and good habits of project-based memory, so your context stays useful instead of bloated.
Basic Token Management
Practical habits for staying under usage limits and keeping sessions efficient. Adds the biggest lever most people miss: don't enable tools, skills, or plugins you aren't actually using, because each one consumes context. Closes on the standard that runs through the whole program: the real work is testing and validating that you understood what you produced.
Level 1 Capstone: Build a ProjectCapstone
Put Level 1 together by building a working Claude Chat project from scratch: system instructions, reference files, and prompts that produce reliable output. You'll use the prompt engineer to refine your setup and then defend the choices you made. Passing means the project works and you can explain why it works.
Getting Started with Skills
Intermediate fluency. This is where you stop using what's built for you and start building for yourself: skills, your own persistent context, and the discipline of managing what the model is holding.
How to Avoid AI Slop
The quality bar for everything that leaves your hands, made explicit. Turns one expectation into a rule: read and edit every AI output before it goes anywhere. AI drafts, you think. A draft you won't take the time to review and revise is not a draft anyone else should have to read. This course sets the standard the rest of the program enforces.
MCP Tools in Practice
What MCP connectors actually do, how to use them in real work, and where the risks live. Connecting Claude to tools like Monday.com, HubSpot, Outlook, and SharePoint, and the discipline of connecting only what a task needs. Reinforces the AI-risk and data-handling principles from Level 1.
Skills 101: The DWM Skills Library
An orientation to skills and the firm's shared library. What a skill is, how it changes what Claude can do, and which DWM skills already exist so you use what's built instead of rebuilding it. How to find, enable, and run a skill on a real task.
Skills 201: Personal Context Portfolio
Build your own Personal Context Portfolio skill: a persistent record of your role, voice, projects, and preferences that Claude loads so you stop re-explaining yourself every session. How to structure it, keep it current, and call it across chats and Cowork.
Context Engineering
The advanced discipline of managing what the model is holding at any moment. Markdown working docs for very long chats; research plans before deep research; knowing when to break a deep dive into a separate chat; how tool and plugin choices affect context; and the right build order for artifacts: model in markdown first, then move to HTML as the final step.
Level 2 Capstone: Build a SkillCapstone
Build and validate a skill that does real work for your role, wired to your Personal Context Portfolio and any tools it needs. You'll design it, test it against real inputs, and show it produces output you would put your name on. Grading is on whether it works and whether you can prove it.
Cowork and Automations
Intermediate-to-advanced fluency. You move from single chats to Cowork, learn to produce artifacts that hold up, and build automations that run real recurring work.
Claude Cowork 101
Getting started in Cowork without over-engineering it. When a task belongs in Cowork, and how to hand off from chat cleanly using a Cowork handoff skill. Deliberately skips plugins and sub-agents to keep the focus where it belongs at this stage: moving work from chat to Cowork efficiently.
Working with Artifacts
How to produce artifacts (documents, dashboards, interactive HTML) that actually hold up. Applies the build order from Context Engineering: draft and model the content in markdown, get it right, then move to HTML as the final step. Sets up the design work in Level 4.
Claude Cowork 201
The next layer of Cowork: scheduled and recurring tasks, and running longer multi-step work reliably. How to set up a task that runs on its own, what to check before you trust it, and how to keep a scheduled job from drifting over time.
Claude Plugins
What plugins are, how they bundle skills for a role, and how to install and use them, including DWM's own C-Suite plugin. Reinforces the context discipline from earlier levels: enable the plugin for the job in front of you, and turn off what you don't need.
Level 3 Capstone: Build an AutomationCapstone
Build a Cowork automation that runs a real recurring task for your role end to end, including a scheduled trigger and a clean chat-to-Cowork handoff. You'll show it runs on its own and produces output that meets the quality bar. Grading is on reliability and your ability to explain how it works and where it could fail.
Advanced Skills
Advanced fluency, with a deliberate focus on avoiding sycophancy and AI slop. You produce polished design work and sharpen your prompting until you can make the model argue with you and grade its own output.
Claude Design
Using Claude Design to produce polished visual work, slide decks first. Builds on the artifacts training to refine design through a structured two-part (sometimes four-part) process: get the content and structure right, then iterate on the design itself. The output is client-grade and internal-grade decks, on brand.
Advanced Prompt Engineering
The advanced layer, and a deliberate refresher on the fundamentals. Pressure-testing your prompts and skills, including Devil's Advocate techniques that force the model to push back instead of agreeing with you. Designing skills directly into project-level prompts. Building your own grading and validation systems so you can check output at scale. It closes the program where it started: engage your own brain, and don't ship slop.
Recommended Resources
Anthropic Learning Resources
Free courses and guides from the makers of Claude, from everyday fluency to building with the API. A natural companion to Levels 1 and 2.
anthropic.com/learn →The Executive Agent Leadership Course
A leadership-level program on directing agents and agentic systems, well suited to the owners and executives in this room.
executiveagentleadership.com →