1 What AI Is

What AI Actually Is

Before you can work effectively with AI, you need to understand what it is. Not the marketing. Not the hype. The actual technology.

01. The Core Idea

What people think AI is

A magical brain in a computer
A search engine that answers questions
Something that "knows" things like a person
Always right, or at least always trying to be

What AI actually is

A pattern-matching engine trained on text
A tool that predicts useful responses to prompts
Something that generates, never retrieves
Confident regardless of whether it is right

AI does not "know" things the way a person does. It generates responses based on patterns in its training data. It is not remembering a fact. It is constructing what looks like a correct answer based on similar patterns it has seen.

02. The Three Ingredients

Training Data

Billions of documents, code, conversations. The raw material AI learned from. It does not "reference" this data at runtime. The patterns are compressed into the model itself.

The Model

A mathematical representation of all those patterns. Think of it as a giant map of how language works. The model is static after training. It does not learn from your conversations.

Inference (Your Prompt)

When you type a prompt, the model runs a prediction to generate a response. This is not "search" or "retrieval." It is generation from statistical patterns, every single time.

03. What This Means for You

The upside

AI can produce novel combinations, new ideas, creative approaches
It adapts to your style, your context, your tone
It can reason through problems step by step
It is tireless and never gets frustrated

The downside

AI can be confidently wrong (hallucination)
It has no internal truth check. It sounds convincing either way
It does not know when it does not know
It will tell you what sounds right, not what is right

💡

Key insight: AI is a reasoning tool, not a knowledge base. It is better at "how should I think about this?" than "what is the answer?" The moment you treat it as an oracle, you introduce risk. The moment you treat it as a thinking partner, you unlock value.

04. What AI Is NOT

Common misconceptions

Not a search engine. It does not retrieve facts from the internet
Not a database. It does not store or recall specific records
Not conscious. It has no feelings, opinions, or self-awareness
Not magic. It is math and statistics applied to language

What it actually is

A pattern-recognition engine trained on text
A reasoning tool that follows instructions
A tireless assistant that never gets bored
A skilled collaborator on first drafts and exploration

The Intern Analogy

Think of AI as a brilliant intern who just joined your team. They have read every book in the library (training data) but have no idea how your specific company works. They need clear instructions, examples of good work, and someone to check their output before it goes to a client. The intern is fast and never sleeps. But they still need you to be the editor.

2 The Shift

How AI Changed (And Why It Matters)

AI did not suddenly appear. It evolved through distinct phases. Understanding these phases explains why today's AI works differently than what you may have tried before.

01. The Three Phases of Modern AI

Phase 1: Rules + Search (Pre-2022)

Old AI was rigid. Chatbots followed decision trees. Search engines ranked keywords. Predictable but brittle. If you asked a question slightly outside their rules, they broke.

Phase 2: The Breakthrough (Late 2022)

Large Language Models. Trained on massive text, AI learned language on its own. Suddenly it could hold a conversation, write, explain, summarize. Flexible in a way old AI never was.

Phase 3: From Chat to Work (2024-2025)

AI moved from answering questions to doing work. Reading files, writing to folders, executing multi-step tasks. The interface changed from chat box to workspace. Most people have not absorbed this yet.

The breakthrough was not that AI got "smarter." It was that AI became flexible. Instead of needing a rule for every input, it could infer what you wanted from patterns in the conversation.

02. Old Mental Model vs. New

Old mental model

Chat with AI like a smarter Siri
Ask one question, get one answer
Copy/paste results manually
No context carried forward
AI is a tool you use occasionally

New mental model

Delegate work to AI like a teammate
Set context, iterate, verify
AI reads and writes files directly
Context files persist across sessions
AI is a collaborator you work with daily

03. Why This Confuses People

The technology moved faster than our mental models. Most people's first experience was ChatGPT in late 2022. So they have spent years thinking of AI as "a thing you chat with." But the technology underneath changed dramatically. The interface looks the same. The capability is fundamentally different.

⚠️

The gap is dangerous: Your mental model of AI is probably outdated if you have not used it in the last 6 months. Treating today's AI like yesterday's chatbot means leaving 80% of its value on the table. Worse, you might dismiss the whole category based on old assumptions.

The bottom line: Every 12-18 months, the capability shifts meaningfully. The specific tools and models will keep changing. But the skills you learn here, context, delegation, verification, collaboration, will apply regardless of what comes next.

3 What AI Is Good At & Not

Strengths & Limits

Effective use of AI starts with knowing what to give it and what to keep. This is not about capability tiers. It is about fit.

01. What AI Excels At

Summarization

Condensing long documents, meetings, transcripts into clear, structured summaries with key points and action items.

Writing & Drafting

First drafts, emails, proposals, reports, content. Give it context and it produces a strong starting point to refine.

Analysis & Comparison

Comparing options across criteria, finding patterns in text data, identifying gaps or inconsistencies in documents.

Brainstorming

Generating ideas, frameworks, approaches. AI is excellent at producing many options quickly for you to evaluate.

Formatting & Translation

Restructuring information, changing tone, translating between languages, turning notes into polished formats.

Research Assistance

Summarizing known topics, explaining concepts, providing overviews. Use it to get oriented, not as a source of truth.

02. What AI Is Bad At

Accuracy Without Verification

AI will confidently state false information. It does not know when it is wrong. Everything must be checked.

Real-Time Facts

Unless connected to search, AI knowledge has a cutoff date. It does not know what happened yesterday.

Math & Calculations

Surprisingly bad at arithmetic and precise math. Always verify numbers, calculations, and data.

Proprietary Knowledge

AI does not know your internal data, customer history, or company processes unless you provide them.

Subjective Judgment

AI cannot make value judgments about your brand, your relationships, or your strategic decisions.

Consistency Over Time

AI may give different answers to the same question. It is probabilistic, not deterministic.

03. The Critical Distinction

The most important skill is knowing which tasks go where. Not every task belongs in AI. Not everything needs a human. The art is recognizing the fit.

Give to AI

First drafts and rough cuts
Summaries and compilations
Comparisons and analysis
Formatting and restructuring
Brainstorming and exploration
Repetitive writing tasks

Keep for yourself

Strategic decisions and direction
Final approval and sign-off
Relationship-sensitive communication
Anything requiring personal judgment
Verification of AI output
Creative vision and taste

A simple rule: AI drafts. You decide. AI proposes. You dispose. AI explores. You choose. The division is not about capability. It is about ownership.

04. The Competency Map

Task Type	AI Strength	Human Requirement
Summarize a meeting transcript	High	Verify key decisions, add context
Draft a client email	High	Review tone, personalize, approve
Calculate quarterly numbers	Low	Use a spreadsheet, verify formula
Research a competitor	Medium	Cross-check facts, add insider knowledge
Write a strategic plan	Medium	Define strategy, review for alignment
Decide whether to fire a client	Low	Human judgment only

Quick test

Before you send a task to AI, ask yourself: would I be comfortable letting someone else make the final call on this? If yes, AI can own the draft. If no, you need to own the decision and use AI for support only.

4 How to Think About New Tools

How to Think About
New Tools

New AI tools and models launch constantly. The framing you use to evaluate them matters more than the features themselves.

01. The Fundamental Question

Every new AI announcement triggers the same cycle: hype, panic, confusion, "will this replace my job?" The useful question is not "is this the best AI?" The useful question is: what does this tool change about how I work?

New models are not new categories. A faster, smarter version of Claude does not change the collaboration pattern. It just means Claude can handle more complex context and follow instructions more precisely. The skill of working with it stays the same.

02. The Evaluation Framework

When a new tool or model appears, run it through these three questions:

Does it change the interface?

Is this still "I give instructions, AI produces output, I verify"? Most tools do not change this. If the interaction model is the same, the skills transfer.

Does it access new data?

Can it read your files, search the web, query your database? This changes what you can give it, not how you work with it.

Does it change who verifies?

This is the critical question. If the tool makes autonomous decisions without human review, that is a real shift. Most tools still put you in charge.

📋

Rule of thumb: If the answer to question 3 is "you still verify," then your existing skills transfer directly. The tool is just a faster version of what you already know. Do not panic. Do not rebuild your workflow. Just test, adapt, and move on.

03. What Actually Changes vs. What Does Not

What changes with each new model

Speed. Responses get faster
Context capacity. Can handle more input
Instruction following. More precise adherence
Reasoning quality. Better logic and analysis
Fewer hallucinations. But never zero

What does NOT change

You still provide context and instructions
You still verify the output
You still apply your judgment
You still own the result
You still need to know what you want

04. The Trap of "Just Wait for the Next One"

A common reaction to fast-moving technology is paralysis: "why learn today's tool when next month's will be better?" This is a mistake. The skills you build now, giving clear context, iterating on output, verifying results, applying judgment, transfer to every future tool.

What compounds: The ability to think clearly about what you want, express it precisely, evaluate what you get, and guide it toward what works. Those skills do not become obsolete. They become more valuable as the tools get faster.

"The best time to start working with AI was two years ago. The second best time is today. Waiting for the next model is like refusing to learn email because something better might come along."

5 The Human Role

The Human Role

AI does not replace judgment. It exposes it. The more AI does, the more your taste, your standards, and your decisions matter.

01. You Are the Editor

Every AI output is a draft. Nothing leaves your desk without your review. This is not a bug. It is the design. AI handles the labor of generation. You handle the labor of judgment. That division is not diminishing your role. It is clarifying it.

AI removes the friction of "getting words on the page." But it does not remove the responsibility of deciding whether those words are right. Your job shifts from writing to editing, from producing to curating, from creating to deciding.

02. What Only You Can Do

🧠

Know Your Context

Relationships, history, politics, unspoken rules. AI does not have any of this.

🎯

Set the Standard

You decide what "good" looks like. AI can match your bar but cannot set it.

✅

Own the Outcome

When it goes out under your name, it is yours. AI does not take responsibility.

⚖️

Make the Call

Strategic decisions, ethical calls, relationship choices. Only you.

03. The Taste Gap

As AI gets better at producing competent output, the differentiator becomes taste. Two people can give AI the same task. One gets something mediocre because they accept the first output. One gets something excellent because they know what to ask for, what to reject, and what to refine. The difference is not AI skill. It is taste.

"AI raises the floor for everyone. But it does not lower the ceiling. The ceiling is your judgment, your standards, your ability to say 'this is not good enough yet' and know exactly why."

04. The Trust Paradox

People new to AI tend to either trust it too little or trust it too much. The right relationship is trust but verify. Trust AI to handle the heavy lifting. Verify everything before it goes out the door.

Trust too little

Ignore AI because it is "not perfect"
Miss the 80% time savings on drafts
Do everything from scratch
Get outpaced by peers who use it

Trust too much

Send AI output without review
Assume AI is correct because it sounds confident
Skip verification steps
Send something that is wrong or off-brand

The right balance

Let AI generate. Review everything before it ships
Trust AI for drafts, research, exploration. Verify facts, tone, and alignment
Use AI to augment your judgment, not replace it

💡

Your superpower: The human ability to say "this is wrong" or "this could be better" is more valuable than ever. AI produces volume. You produce judgment. That combination beats either alone.

6 How to Read AI Output

How to Read
AI Output

The single most underrated AI skill is not prompting. It is reading. Knowing what to look for determines whether you catch errors or miss them.

01. The Fluency Problem

AI output reads fluently. That is the problem. Bad AI output does not look wrong. It looks perfectly reasonable. The sentences are grammatical. The structure is logical. The tone is confident. This fluency makes us lower our guard. But AI does not know what it is talking about. It sounds like it does.

Fluency is not accuracy. The most dangerous AI output is the kind that reads perfectly but is completely wrong. You have to read with suspicion, not with trust.

02. The Two-Pass Reading Method

Pass 1: Substance Scan

Read for facts, claims, numbers, names, dates. Does anything seem off? Flag everything questionable before refining.

Pass 2: Fit Check

Read for tone, audience, purpose. Does this match the brief? Does it sound like your company? This pass is about alignment, not correctness.

📋

Order matters: Always check substance before style. It is easy to get pulled into polishing a paragraph that should not exist in the first place. Facts first. Then fit.

03. Common Failure Modes

Hallucination

AI states false information confidently. Names, dates, events, statistics. These look real but are invented. Always verify specific claims.

Sycophancy

AI agrees with you even when you are wrong. If you ask "isnt this the right approach?" AI will likely say yes. Push back and ask it to challenge you.

Oversimplification

AI reduces complex topics to surface-level summaries. It misses nuance, trade-offs, and edge cases. Dig deeper on complex subjects.

False Specificity

AI invents specifics to sound authoritative. "According to a 2023 Gartner report..." when no such report exists. Watch for plausible-sounding citations.

04. The Verification Checklist

Before you use any AI output, check:

Are the facts verifiable? (names, dates, numbers, citations)
Does the tone match the audience and purpose?
Are there any claims that seem too good (or bad) to be true?
Does it miss important context that only you know?
Would you be comfortable putting your name on this?

Example

Claude writes: "According to a 2024 McKinsey study, companies using AI saw a 40% increase in productivity." Sounds credible. Sounds specific. But is it real? If you cannot verify the source, do not use the claim. Always ask Claude for specific, verifiable citations on factual claims.

7 AI Safety & Privacy

AI Safety & Privacy

Using AI effectively means using it responsibly. Understanding what data goes where, and what to keep private, is not optional.

01. The Core Rule: What You Put In Matters

Everything you type into an AI tool is processed by that company's servers. This includes prompts, documents you upload, and any personal or proprietary information you include. The general rule: do not put anything into an AI tool that you would not put in an email to a contractor.

⚠️

Common mistake: People paste confidential client data, internal financials, personal health information, or proprietary code into AI tools without thinking. Treat prompts like you would treat any external communication.

02. Data Handling by Platform

Platform	Data Use Policy	Key Consideration
claude.ai (Free/Pro)	Not used for training by default	Anthropic does not train on your API or claude.ai data. You can opt in to share for improvement.
Claude API	Not used for training	Data is processed but not retained for training. Enterprise-grade data handling.
Claude Team/Enterprise	Not used for training	SOC 2 compliant. Data never leaves your trust boundary. Includes admin controls.
ChatGPT (Free/Plus)	Used for training unless opted out	By default, conversations may be used for training. Opt out in settings. Team/Enterprise versions exclude training use.

When in doubt about data sensitivity, use the enterprise tier or API, or skip AI entirely for that task. No productivity gain is worth a data breach.

03. Privacy Best Practices

Do put in

Public information and research
Your own writing and notes
Anonymized scenarios and data
General business context (without names)
Things you would put in a shared doc

Do NOT put in

Passwords, credentials, API keys
Personal health information (PHI)
Customer PII (names, emails, addresses)
Confidential financial data
Trade secrets or proprietary code

04. The Hallucination Risk (Revisited)

Hallucination is not just a quality issue. It is a safety issue. If AI invents a statistic that makes it into a client presentation, or fabricates a legal precedent that gets cited in a contract, that is a real-world harm.

📋

High-stakes rule: For anything that goes to a client, a regulator, or the public: (1) Draft with AI. (2) Fact-check every specific claim. (3) Have a human review the final version. (4) Never skip step 3.

05. The Trust Boundary

Know your company's policy

Your company may have specific rules about which AI tools are approved and what data can be shared.

Check before sharing client data

If you handle client information, verify whether your AI tool is approved for that level of data.

Use enterprise tiers for sensitive work

Free tiers do not offer the same data protections as enterprise or API-based access.

When in doubt, leave it out

If you are unsure whether data is safe to share, do not share it. Anonymize or skip AI for that task.

Safety and privacy are not constraints on AI use. They are the foundation that allows responsible AI use. Get this right first, then focus on productivity.

8 The Collaboration Pattern

The Collaboration
Pattern

This is the pattern that ties everything together. A repeatable loop for working with AI, regardless of tool, model, or task.

01. The Core Loop

Every interaction with AI follows the same four-step pattern. Master this pattern and you can work effectively with any AI tool.

Orient

Set context. Who are you? What are you trying to do? What does success look like? This step determines everything that follows.

Direct

Give clear instructions. What role should AI take? What format? What constraints? Vague direction produces vague results.

Review

Read the output critically. Check facts, tone, alignment. Does this meet the bar? If not, what needs to change?

Iterate

Refine. Tell AI what to adjust. Give feedback on what worked and what did not. Stop when it is good enough, not when it is perfect.

The loop applies at every scale. A single email. A 50-page proposal. A brainstorming session. Orient. Direct. Review. Iterate. Every time.

02. Surface-Level vs. Deep Collaboration

Surface-level

One-shot prompts. Accept first output
No context provided upfront
Vague instructions ("make it better")
No iteration or refinement
Starts fresh every time

Deep collaboration

Rich context and examples upfront
Clear success criteria defined
Multiple rounds of refinement
Feedback that teaches AI your preferences
Builds on shared context over time

03. The Feedback Loop

AI learns nothing from a single conversation. But you learn something from every interaction. Each time you refine your prompt, each time you catch an error, each time you clarify your instructions, you are building a skill that transfers to every future interaction.

The compounding effect: The first time you use AI for a task, it might take as long as doing it yourself. The third time, half as long. The tenth time, a quarter of the time. The skill is not in the tool. The skill is in your ability to orient, direct, review, and iterate efficiently.

04. Common Anti-Patterns

The Kitchen Sink

Dumping everything into one prompt. AI gets overwhelmed. Break complex tasks into smaller conversations.

The Endless Thread

Keeping one conversation going for weeks. Context window fills. Quality degrades. Start fresh, bring context forward.

The Blind Accept

Accepting first output without review. Misses errors and opportunities for improvement. Always review.

The Vague Direction

"Make this better." AI has no idea what "better" means. Be specific about what you want.

05. The Meta-Skill

The collaboration pattern applies whether you use Claude, ChatGPT, Gemini, or a tool that does not exist yet. The interface will change. The models will get smarter. The pattern will remain the same.

"The tools will evolve. The models will improve. The skills of context, clarity, and judgment will only become more valuable. Invest in the pattern, not the platform."

06. Durable Skills Summary

Skill	Why It Lasts
Context setting	Every AI tool works better with clear context. This will never change.
Instruction clarity	Specific direction beats vague requests regardless of model capability.
Critical reading	AI output always needs human review. The better AI gets, the more subtle the errors.
Iteration	First drafts are starting points. Refinement is where quality comes from.
Verification	Trust but verify. This applies to every AI tool, every model, every output.
Judgment	AI proposes. You decide. That division is permanent.

💡

What's next: The specific training on Claude follows this module. You will learn tool setup, prompting framework, context files, and workflows. The skills here are the foundation. Everything that follows builds on them.

9 Why Set Up Claude

Why Set Up Claude?

AI is everywhere. So why invest time in this tool, this setup, this workflow? Because the difference between "trying AI" and "integrating AI into how you work" is the difference between a toy and a transformation.

08-01 · The Real Cost of Not Setting Up

❌ Without Claude

Tab-hopping between tools, losing context each time
Mental overhead: remembering everything yourself
Starting from scratch on every project
Scattered notes, no shared memory
Repeating the same work daily
No system for leveraging AI output

✓ With Claude Set Up

One workspace for thinking, writing, and creating
Claude holds your project context, you hold the vision
Reusable templates and workflows
Structured context files that persist between sessions
Compound productivity: each session builds on the last
A system that amplifies your thinking

The question isn't "can AI do this?" It's "are you set up to make AI work for you, consistently?"

08-02 · AI as Your Second Brain

Your biological brain is great at creative leaps, intuition, and judgment. It is terrible at holding lots of details, processing large documents, and remembering everything. Claude fills those gaps.

Externalize memory

Instead of keeping project details, requirements, and context in your head, put them in Claude's context window. Free your mental RAM for higher-level thinking.

Think in public

Use Claude as a thinking partner. Draft ideas, get feedback, explore angles. Seeing your thoughts reflected and reshaped by AI clarifies your own thinking.

Process at scale

Claude reads faster than you. Paste a 50-page document, a codebase, a research paper. Ask questions about it. Claude has already read it.

Build a knowledge base

Your project folder becomes your shared context. Documents, notes, examples, and guidelines that Claude references consistently, every single conversation.

Iterate faster

First draft in seconds. First revision in another pass. By the time you'd normally start writing, you're already refining. Speed without sacrificing quality.

🧠

The Second Brain Mindset: Your biological brain is for having ideas. Claude is for processing information. When you stop trying to hold everything in your head and start using Claude as an extension of your thinking, two things happen: your productivity goes up and your mental load goes down. That's the setup payoff.

08-03 · What Setup Unlocks

🧠 Project Context

Your project folder with instructions, examples, and guidelines. Claude reads it every session. No more repeating yourself.

📋 Consistent Output

Templates and patterns that enforce your standards. Same quality, every time, regardless of how you phrase the prompt.

🔁 Reusable Workflows

Save prompt patterns for common tasks. Don't reinvent the wheel for every email, document, or analysis.

🔗 Tool Integration

Claude connects to your codebase, files, and tools. One interface for writing, coding, analyzing, and creating.

⚡

Faster execution

Tasks that took hours take minutes

🧘

Lower cognitive load

Less mental juggling, more clarity

📈

Compound knowledge

Each session builds on the last

🎯

Higher quality

Consistent standards, every output

08-04 · What Setup Is (And Isn't)

Setup is NOT

Hours of configuration before you see value
Learning to code or becoming a prompt engineer
Replacing your judgment with AI
A one-time task you do and forget

Setup IS

A 15-minute investment that pays off every single day
Creating a project folder and a few context files
Learning a delegation framework you can use anywhere
An evolving system that gets better as you use it

Setup is not a technical hurdle. It is a force multiplier. The 15 minutes you invest now will save you hours every week.

10 Opening & Foundations · 7 slides

The Way You Work
Is About to Change

This isn't a technical training. You don't need to know how to code. You just need to know how to delegate.

01-01 · Old Paradigm vs. New Reality

❌ The Old Way

Treat AI like a search engine, ask once, move on
Zero context provided
Treats AI as "magic answer box"
One-shot prompts, no iteration
No follow-through or verification
Starts fresh every single time

✓ The New Paradigm

Delegate like a teammate, set context, iterate, verify
Rich context and examples upfront
Treats AI as a skilled collaborator
Iterative refinement loop
Verify and validate all outputs
Builds on shared understanding

The shift: from asking questions → delegating outcomes

01-02 · Tokens: How AI Reads Your Words

AI doesn't read words like we do. It reads in "tokens", chunks of text about ¾ of a word long.

~¾ of a word

1 token ≈ ¾ of a word in English. "Hello" = 1 token. "Collaboration" = 2-3 tokens.

200,000 tokens

Claude can hold ~200K tokens in memory at once, roughly a 300-page book.

The limit matters

When the limit fills, Claude forgets the oldest context first. This is why long chats degrade.

Different languages

Non-English text uses more tokens per word. Plan accordingly if working multilingually.

01-03 · The Context Window: How AI Remembers

📋

Analogy: Imagine a whiteboard. Everything you write stays there until you erase it. Claude's whiteboard holds ~200,000 tokens, roughly a 300-page book. Everything in the current conversation goes on the whiteboard. When you start a new conversation, you get a fresh whiteboard. Old conversations are NOT remembered unless you bring the info forward.

The context window is your shared workspace. The more relevant information you put in it, the better Claude's output will be.

01-04 · The Beginner Trap: Why Conversations Break

The Trap

You ask Claude something → It works great → You ask something else → Works less well → You add more instructions → Context window fills up → Claude "forgets" the beginning → "Why is it getting worse?"

💡

The solution:
• Long conversation? Start a fresh one with ALL the context re-pasted
• Different task? Start fresh. Don't pile onto old conversations.
• Complex task? Break it into smaller conversations.

01-05 · The 4-Step Delegation Flow

Give context

Background, audience, goal, constraints, before anything else.

Set expectations

Tone, format, length, what success looks like.

Define the task

Clear, specific instructions, not vague requests.

Iterate & verify

Review, refine, check for correctness. First draft = starting point, never final.

Delegating to AI is like delegating to a capable teammate. The more context and clarity you provide upfront, the less rework you'll do.

01-06 · Exercise: Reflect on Your Current Use

✏️

Part 1. Think (2 min): How do you currently use AI? Do you ask single questions and move on? Give context? Iterate?

Part 2. Write (3 min): Write one task you do regularly that you'd like AI to help with more effectively.

Part 3. Share (optional): What's one way your current use could improve with the "delegation" mindset?

11 Claude & The Landscape · 7 slides

Meet Your
New Teammate

Claude is not just a chatbot. In 2026, it's six tools. Most people only use one of them.

02-02 · What Is Claude?

Claude is an AI assistant created by Anthropic, a company founded by former OpenAI researchers. Built for collaboration, not just chat.

📄

200K context

Processes entire documents, books, months of email threads without losing track.

🎯

Instruction-following

Purpose-built to follow detailed, complex instructions precisely. Excels at nuanced tasks.

🖥️

File-aware

Desktop Cowork app reads your actual folder, not just text you paste in.

🛡️

Constitutional AI

Built-in safety guidelines. Designed to push back on problematic requests.

We use Claude because it excels at long context, nuanced judgment, and following detailed instructions.

02-03 · Claude vs ChatGPT: The Honest Comparison

Claude wins at

Larger context window (200K vs 128K tokens)
Following complex, detailed instructions
Nuanced analysis and reasoning
Writing that sounds human, not generic
File work via Cowork
Constitutional AI safety approach

ChatGPT wins at

Voice mode (much stronger)
Image generation (Claude can't)
Creative brainstorming and writing
More third-party integrations
Better multimodal (DALL-E)
Broader plugin ecosystem

💡

Neither is "better", they excel at different things. Most power users use both. What matters more than which model you pick is how much context and clarity you provide.

02-04 · Three Ways to Use Claude

💬

Chat (claude.ai)Start here

Browser-based. Best for quick drafts, questions, brainstorming, research. No setup required. Use it like texting a smart colleague.

🖥️

Cowork ModeKey feature

Desktop app. Works in your actual project folder. Can read, write, and edit files. This is where the biggest productivity jump happens.

⌨️

Claude Code (Terminal)

Advanced: runs in terminal, can execute commands, manage complex projects. Not needed to start, we'll skip for now.

We'll focus on Chat and Cowork Mode. These two cover 95% of what you'll need day to day.

02-05 · What You Can Actually Do

Writing & Communication

Draft and refine emails
Write and edit proposals
Summarize meetings and documents
Create reports and content
Rewrite for different audiences
Proofread and check tone

Analysis & Research

Analyze data and find patterns
Research topics and summarize findings
Compare options and recommend
Review documents for gaps
Extract key info from long texts
Brainstorm and plan projects

If you spend time reading, writing, or thinking at work. Claude can save you significant time.

02-06 · Our Recurring Example: Meeting → Email

We'll use this scenario throughout the entire training to see how each skill applies:

The scenario: You just attended a 45-minute client meeting. Now you need to: (1) write a meeting summary, (2) draft a follow-up email, (3) create an action item list.

Without AI

30-45 minutes of writing, checking, and formatting, from scratch, every time.

With AI (done well)

5-10 minutes of directing, reviewing, and refining, using context you already have.

02-07 · Exercise: What Would You Use Claude For?

✏️

Part 1 (3 min): Look at your calendar from last week. Pick 3 tasks you spent time on, writing, summarizing, researching, planning. Write down how you might use Claude for each one.

Part 2 (2 min): Share one idea with the group. What task would save you the most time?

Check-in: Has anything surprised you so far? What's still unclear?

12 Which AI Tool for What · 7 slides

Which AI Tool for What

There's no single "best" AI tool. Each has strengths. The skill is matching the tool to the task - and knowing when to use more than one.

11-01 · The AI Tool Landscape

The AI space is crowded. New models launch weekly. But the tools worth knowing fall into a few distinct camps. Each excels in different areas. Your goal is not to pick one - it's to know which to reach for based on the task.

ChatGPT Creative

Best for creative writing, voice conversations, image generation (DALL-E), brainstorming, and broad knowledge tasks. Strong plugin ecosystem.

Claude Analysis

Best for long documents (200K context), complex instruction following, nuanced analysis, coding, and file-aware workspace.

Gemini Integration

Best for Google Workspace integration (Gmail, Docs, Drive), video understanding, and research with Google Search grounding.

Specialized Tools

Perplexity (research + citations), GitHub Copilot / Cursor (coding), Midjourney / Canva (image), NotebookLM (research org), ElevenLabs (voice).

The landscape changes fast. What matters is not memorizing features - it's knowing which interaction style fits which task. The skills of context, iteration, and verification transfer across all of them.

11-02 · ChatGPT: Best For

ChatGPT (OpenAI) is the most widely used AI tool. It excels at creative and conversational tasks, and has the broadest feature set including voice, image generation, and a large plugin marketplace.

Strengths

Creative writing & storytelling. Fiction, scripts, ad copy. ChatGPT's creative flexibility is top-tier.
Voice conversations. Advanced Voice Mode is best-in-class. Natural, real-time voice interaction.
Image generation. DALL-E integration. Generate and iterate on images directly in chat.
Brainstorming. Excellent at generating many ideas, angles, and approaches quickly.
Plugin ecosystem. Broadest set of third-party integrations and plugins available.

Limitations

Smaller context. 128K tokens vs Claude's 200K. Can't process entire books or massive docs.
Generic writing. Output can feel more templated and less nuanced than Claude.
Less precise instruction following. Tends to overshoot or miss complex constraints.
No file workspace. Can't read/write your project files without third-party plugins.

11-03 · Claude: Best For

Claude (Anthropic) is built for collaboration on complex work - not just conversation. It excels where precision, nuance, and long context matter.

Strengths

Long document processing. 200K token context window. Read entire books, months of email, massive codebases.
Instruction following. Best-in-class at following detailed, nuanced, multi-part instructions.
Nuanced analysis. Superior at reasoning through complex problems with careful judgment.
Coding & technical work. Excellent at code review, debugging, architecture analysis.
File workspace. Cowork Mode reads/writes your actual project files. No copy-paste needed.
Human-sounding writing. Output reads naturally, not like a template. Less "AI-sounding."

Limitations

No image generation. Claude cannot generate images natively.
Voice is weaker. ChatGPT's voice mode is significantly more advanced.
Fewer integrations. Smaller plugin ecosystem compared to ChatGPT.
Higher learning curve. Claude's full power requires setup (context files, Cowork, skills).

11-04 · Gemini: Best For

Gemini (Google) is deeply integrated into the Google ecosystem. Its strongest use case is working with Google Workspace and multimodal understanding.

Strengths

Google ecosystem. Direct integration with Gmail, Google Docs, Drive, Sheets. Access your existing content.
Multimodal understanding. Best video understanding capability. Can reason about video content and audio.
Web research. Integrated Google Search grounding for real-time, cited information.
Free tier. Very capable free tier. Excellent for casual users and students.

Limitations

Weaker at nuance. Can feel less precise than Claude for complex reasoning tasks.
Less consistent. Output quality can vary more between use cases.
Google dependency. Most powerful features require Google Workspace subscription.
Smaller ecosystem. Fewer third-party integrations and community resources.

11-05 · Specialized AI Tools

Beyond the general-purpose models, specialized tools handle specific tasks better than any generalist. These are worth knowing even if you mainly use Claude or ChatGPT.

🔍 Perplexity

AI-powered search engine. Answers questions with citations from live sources. Best for research, fact-checking, and staying current. Not a general assistant - it's search with reasoning.

💻 GitHub Copilot / Cursor

Code-specific AI. Integrated into VS Code and other IDEs. Best for writing, completing, and debugging code. Not designed for documents, emails, or general tasks.

🎨 Midjourney / Canva

Image generation. Midjourney for artistic, high-quality images. Canva for design templates and marketing graphics. Use these when you need visuals, not text.

📚 NotebookLM

Google's research notebook. Upload documents, it answers questions based only on your sources. Best for deep research on specific materials. No hallucinations outside your sources.

🎤 Granola

AI meeting notes. Records, transcribes, and summarizes meetings automatically. Best for anyone who attends frequent meetings. Saves hours of manual note-taking per week.

🗣️ ElevenLabs

AI voice generation. Best-in-class text-to-speech, voice cloning, and audio content. Use for narration, voiceovers, and audio content. Not for text-based work.

Specialized tools beat generalists at their specific job every time. Use Claude or ChatGPT for thinking and writing. Use Perplexity for research. Use Copilot for coding. Use Midjourney for images. The right tool for the right task.

11-06 · The Power User Strategy

Power users don't pick one AI tool. They use 2-3 tools as a system, each for what it does best. The skill is not knowing every feature - it's knowing which tool to reach for and when.

✓ Power user workflow

Claude for analysis, long documents, writing, coding, file work. Your main workspace.
ChatGPT for creative brainstorming, voice conversations, image generation. Your creative partner.
Perplexity for research, fact-checking, current events. Your research assistant.
Gemini for Google Workspace tasks, working with your existing Google docs and emails.

The principle

Use the right tool for the task, not one tool for everything
The skills you're learning here (context, iteration, verification) transfer to every tool
Don't chase new models - chase better workflows
The tool that's best today won't be best in 6 months. The skill of working with AI will still matter.

The meta-skill is not "learning Claude" or "learning ChatGPT." It's learning how to work with AI in general - context, iteration, verification - and applying it to whatever tool fits the moment.

11-07 · Exercise: Match the Tool

✏️

For each scenario, pick the best tool(s) and explain why:

1. You need to analyze a 150-page legal contract and draft a summary with risk flags.
2. You want to generate a hero image for a landing page and brainstorm taglines.
3. You need to research a competitor's recent funding round (announced yesterday) and summarize the key details.
4. You're drafting a client email and need it to sound exactly like you - your tone, your vocabulary, your style.
5. You have 50 files in Google Drive that need to be organized and summarized for a quarterly review.

Discuss (2 min): Share your answers with a partner. Did anyone pick differently? What drove the choice?

💡

Answers: (1) Claude - 200K context handles 150 pages easily. (2) ChatGPT - DALL-E for image + brainstorming. (3) Perplexity - live search with citations. (4) Claude - most precise instruction following + file workspace for context files. (5) Gemini - native Google Drive integration.

13 Prompting · 10 slides

How to Talk to Claude
So It Actually Works

Prompting is not about finding "magic words." It's about clearly communicating what you need to a capable teammate.

03-02 · Delegation, Not Magic

If you asked a new team member to write a client email, you'd tell them: who the client is and what they care about, what the email should accomplish, what tone to use, what to include and avoid. The same rules apply to prompting AI.

Prompting is a conversation skill, not a technical skill. If you can explain a task to a colleague, you can prompt Claude.

03-03 · The 6-Part Prompt Framework

1. Role

"You are a senior copywriter...", gives Claude an expertise lens. Fundamentally changes the output quality.

2. Context

Background, audience, situation. Claude has no idea who you are. Your job: set the scene.

3. Task

What exactly should it do? Be specific. "Make it better" tells Claude nothing.

4. Constraints

Length, tone, what to avoid. Set the bounds so Claude doesn't overshoot.

5. Output Format

Email? Bullet points? Table? Memo? Specify it or Claude will choose for you.

6. Verification

"Flag anything that sounds like a promise we haven't approved." Tell it how to check its own work.

Not every prompt needs all 6, but the more you include, the better the result. Aim for at least 3-4.

03-04 · Rule 1: Give Context (The Most Important Rule)

Why it matters: Claude has no idea who you are, who you're writing to, or what's happened before. Your job is to set the scene.

❌ Weak, no context

"Write an email to a client."

Claude guesses the client, purpose, and tone. Almost always generic and wrong.

✓ Strong, rich context

"Write an email to Sarah Chen at Acme Corp. We met yesterday to discuss their Q3 website redesign. She expressed concern about timeline. Purpose: reassure her and confirm next steps. Relationship: friendly but professional."

Context is the difference between "I'm here" vs "I'm at the coffee shop on 3rd Street." Same message, completely different usefulness.

03-05 · Rule 2: Give Examples

Claude learns from patterns. An example is worth 100 words of explanation.

❌ No example

"Write a professional bio."

→ Generic corporate bio that sounds like everyone else.

✓ With example

"Write a professional bio. Here's one I like, notice the tone: [example]. Write mine in the same style."

→ One example replaces paragraphs of instruction.

Examples are the fastest way to communicate tone, style, quality, and format. One good example replaces paragraphs of instruction.

03-06 · Bad Prompt → Good Prompt

❌ Bad prompt

"Write a blog post about AI."

No audience · No tone · No structure · No examples · No constraints · No format

✓ Good prompt

Role: Technology writer for B2B audience Context: Readers are ops leaders skeptical about AI Task: 500 words. 3 concrete ways AI saves time in daily ops Tone: Direct, evidence-backed, no hype Format: Headline + 3 sections with bullets. End with 1-sentence takeaway Verify: Flag any overgeneralizations

The good prompt takes 30 seconds longer to write. It saves 30 minutes of back-and-forth editing.

03-07 · Let's Build a Prompt Together (Recurring Example)

Remember our scenario: meeting recap → follow-up email. Here's a prompt built step by step:

Role: You're a client success manager who attended a 45-min project review with Acme Corp. Context: The meeting covered Q3 deliverables, timeline concerns, and budget approval. The client seemed satisfied but wanted clearer milestones. Task: Write a meeting summary with key decisions, action items, and a follow-up email to the client. Constraints: Keep the summary under 300 words. The email should be warm but professional. Do not assume commitments we haven't discussed. Format: Summary first (bullet points), then email draft below. Verify: Check that all action items are specific and assigned. Flag anything that sounds like a promise we haven't approved.

03-08 · Practice: Rewrite a Weak Prompt

Exercise

Here's a weak prompt. Let's improve it together:

"Write a status update for the project."

✏️

Questions to ask yourself:
• What's missing? (Context, audience, format, tone)
• What would make this specific? (Which project? To whom? How much detail?)
• What should it include? (Progress, blockers, next steps?)

The rewrite should include: Role, Context (what project phase), Task (inform stakeholders or request input?), Format (email, doc, slide), Constraints (length, level of detail).

03-09 · Quick Win: Summarizing With Claude

This is one of the easiest ways to start. Paste any text, meeting notes, article, email thread, and use:

Summarize this in 3-5 bullet points. Include key decisions, open questions, and action items.

Paste

Copy the text (meeting transcript, article, email thread)

Prompt

Use the template above, or customize it.

Refine

"Shorten to 2 sentences" / "Format as email" / "Translate for a non-technical audience"

Summarization is Claude's superpower. The 200K context window means it can handle entire books, legal documents, or months of email threads.

03-10 · Exercise: Write Your First Real Prompt

✏️

Part 1 (1 min): Pick a real task from your work this week, an email, a summary, a draft.

Part 2 (5 min): Write a prompt using the 6-part framework: Role, Context, Task, Constraints, Format, Verify.

Part 3 (optional): Run your prompt. Review the output. What would you change?

Check-in: Which part felt most useful? What was hard about writing the prompt?

14 Your Project Folder · 4 slides

Your Project Folder

5 minutes of setup saves you hours of re-explaining yourself every session. This is the foundation of everything.

04-02 · Why Work in a Folder?

If you use Claude on the website (claude.ai), every conversation starts completely blank. Claude has no idea about your work, your clients, or your context.

Without a folder

Every conversation starts blank
Claude has no idea who you are
You re-explain context every single time
Nothing is saved or organized
Like asking a stranger on the street

With a folder

Claude reads your files = instant context
Claude writes and updates files = work saved
Everything organized, nothing lost
Context files live here permanently
Like handing a dossier to a briefed teammate

The folder is your shared workspace. All your context lives here. Claude reads from it, writes to it, and every conversation builds on what's there.

04-03 · Setting Up Your Project Folder

Create the main folder

On Desktop or Documents, create: Claude Cowork

Create subfolders inside it

ABOUT ME, your context files
OUTPUTS. Claude saves all work here
TEMPLATES, reusable prompts you build over time

Download Claude desktop app

Go to claude.com/download. Install it. This unlocks Cowork Mode.

Open Cowork → New Task → Select folder

Claude is now connected to your workspace and can see all files inside it.

Select model: Opus 4.6 + Extended Thinking

Use the dropdown. Extended Thinking = Claude reasons step-by-step before answering.

Folder structure

📁 Claude Cowork/ ├── 📁 ABOUT ME/ ← your context files live here │ ├── about-me.md ← who you are, how you work │ ├── claude.md ← standing orders for every session │ └── anti-ai.md ← writing patterns to avoid ├── 📁 OUTPUTS/ ← Claude saves all work here └── 📁 TEMPLATES/ ← reusable prompts and examples

Run this every time you start a new Cowork session:

Read every file in the ABOUT ME folder. Tell me what you know about me and what instructions you'll follow. If anything is unclear, ask me.

04-04 · Exercise: Create Your Folder Now

Created Claude Cowork folder on Desktop or Documents
Created ABOUT ME subfolder inside it
Created OUTPUTS subfolder inside it
Created TEMPLATES subfolder inside it
Downloaded Claude desktop app from claude.com/download
Opened Cowork Mode, selected the Claude Cowork folder

15 Context Files · 10 slides

The Files That Make
Claude Actually Yours

Write these once. Use them forever. This is the difference between "generic AI output" and "this sounds exactly like my work."

05-02 · Why Context Files Matter

Without context files

"Who are you?" → Claude doesn't know
"Who are you writing to?" → Claude doesn't know
"What's your style?" → Claude guesses
"What's happened before?" → Claude doesn't know

With context files

Claude knows your role, company, audience
Claude knows your writing style and preferences
Claude knows your past decisions and guidelines
Claude knows your common tasks and templates

Context files are a "briefing document" you hand to Claude at the start of every conversation. Write them once. Use them forever.

05-03 · The 3 Essential Context Files

about-me.mdMost important

Who you are, your role, your communication style, your preferences. Like a personal briefing document.

writing-rules.md

Your writing guidelines: tone, formatting rules, words to use and avoid, style preferences.

decision-rules.md

Past decisions, common scenarios, rules of thumb. "If X happens, we do Y." Prevents repeated mistakes.

Start with one file. Add more as you discover what Claude keeps getting wrong about you.

05-04 · Template: about-me.md

about-me.md, example output

# About Me ## My Role Senior Account Manager at [Company] ## My Communication Style - Direct but warm - Prefer bullet points over paragraphs - Use "we" not "I" whenever possible - Avoid jargon, plain language ## My Preferences - I review everything before sending - I prefer options, not single recommendations - Flag risks and concerns before benefits ## What I Value - Clarity over cleverness - Specificity over generalization - Actionable over theoretical ## Instructions for Claude 1. Read this file before starting any task 2. Ask me clarifying questions before executing 3. Save all outputs to OUTPUTS/ folder

💡

Best way to create it: Let Claude interview you, it asks questions one at a time, then writes the file for you. Pick the approach that fits: Short (practical, 20 questions, under 2K tokens) or Long (deep voice capture, 100 questions, full reference doc).

Short, quick setup, functional profile

You are building my about-me.md file for my Cowork folder. This file will be read by Claude at the start of every session to help you do my job with me. It needs to be concise and high-signal. Your job: interview me using AskUserQuestion (20 questions), then compile the answers into a condensed about-me.md under 2,000 tokens. ## How to interview me Use AskUserQuestion for every question. One question at a time. Let me use "Other" to dictate long answers when I need to. If I give a vague answer, push back. Ask for a specific example or rephrase. Don't accept "I like to keep things clear" without knowing what clear looks like in my work. Follow interesting threads. If something unexpected comes up, go deeper before moving on. ## What to cover (15-20 questions, adapt based on what matters for my role) WHO I AM (3 questions) - What do I do? What's my role, my company, my industry? - Who do I work with or work for? (clients, team, stakeholders, audience) - What does a good week of work look like for me? HOW I WORK (4 questions) - What tools do I use every day and how? - Walk me through how I start a typical task from zero to done. - What does my review/editing/QA process look like? - When I hand something off (to a client, a boss, a reader), what does "done" look like? WHAT GOOD LOOKS LIKE (4 questions) - Show me or describe the best output you've produced recently. What made it good? - What separates great work from average work in your field? - When you look at someone else's work and think "this is good," what are you reacting to? - If I had to judge your work, what should I be looking for? WHAT YOU HATE (4 questions) - What's an example of bad work in your field? What specifically makes it bad? - What patterns, shortcuts, or habits in your industry make you cringe? - When Claude writes something for you and it's wrong, what's usually off? (tone, structure, detail level, assumptions) YOUR RULES (3 questions) - What do you never do in your work? Hard lines you won't cross. - What are the 2-3 non-negotiables that every piece of your work must have? YOUR OPINIONS (3 questions) - What do you believe about your field that most of your peers would push back on? - What tools, methods, or trends do you think are overrated? What's underrated? ## Output format After the interview, compile everything into a single markdown file. Do NOT save raw Q&A transcripts. Extract the patterns from my answers and write them as condensed prose and bullet points. Structure: # ABOUT ME: [My Name] ## Who I am [2-3 sentences. My role, my work, my audience/clients. Current facts and numbers if relevant.] --- ## How I work [My daily tools, my process, how I start tasks, how I review, what "done" looks like. Short paragraphs.] --- ## What good looks like [What I value in my own work and others'. The standards I hold. Condensed from examples I gave.] --- ## What I hate [Patterns, shortcuts, and mistakes that bother me. What "wrong" looks like. Specific, not vague.] --- ## My rules [Numbered list. Hard lines and non-negotiables.] --- ## Instructions for Claude [10 numbered rules for how to work with me. Derived from everything above. Focus on what Claude must DO and NOT DO, not abstract principles.] Target: under 2,000 tokens total. Every sentence should carry signal. If a sentence could be cut without losing information, cut it. Save the file as about-me.md in my ABOUT ME/ folder.

Long, deep voice & taste capture, full reference

You are a Taste Interviewer. Your job is to extract the DNA of how I think, write, and see the world, so precisely that another Claude instance could write exactly like me. Conduct 100 questions across: beliefs & contrarian takes (15), writing mechanics (20), aesthetic crimes (15), voice & personality (15), structural preferences (15), hard nos (10), red flags (10). Rules: one question at a time. Push back on vague answers. Ask for specific examples. Call out contradictions. Follow interesting threads. After 100 questions, compile into a comprehensive markdown document with sections for each category, a quick reference card, and an anti-overfitting guide. Preserve the full depth of every answer, this is a reference document, not a summary.

05-06 · Template: writing-rules.md

writing-rules.md

# Writing Rules ## General - Use active voice ("We recommend..." not "It is recommended...") - Keep sentences under 25 words where possible - One idea per paragraph ## Words to Avoid - "Leverage" → Use "use" instead - "Synergy" → Use "partnership" or "collaboration" - "Best-in-class" → Show don't claim ## Formatting - Use bullet points for lists of 3 or more - Bold for emphasis, not decoration - Headlines: sentence case, not title case

05-07 · Template: decision-rules.md

decision-rules.md

# Decision Rules ## Pricing Questions - Never share pricing without a proposal - Always flag discount requests for manager approval ## Client Communication - Always address concerns within 24 hours - Copy account manager on all escalation emails - Use client's preferred communication channel ## Content Approval - Social posts: team lead approval, 24-hour turnaround - Blog posts: content lead + legal review - Client deliverables: project lead sign-off only

05-08 · Template: anti-ai.md

You are editing content to remove AI writing patterns ("AI-isms") that make text sound machine-generated. This skill operates in two modes: rewrite (default) or detect. In rewrite mode: audit the text, rewrite it clean, show a diff summary. In detect mode: audit the text, flag issues but don't rewrite. ## What to flag and fix ### Formatting - **Em dashes**: Replace with commas, periods, parentheses, or split into two sentences. Max one per 1,000 words. - **Bold overuse**: Strip bold from most phrases. One bolded phrase per major section max. - **Emoji in headers**: Remove entirely. No "## 🚀 What This Means". - **Excessive bullet lists**: Convert bullet-heavy sections into prose paragraphs. ### Sentence structure - **"It's not X, it's Y"**: Rewrite as a direct positive statement. Max one per piece. - **Hollow intensifiers**: Cut "genuine," "real" (as in "a real improvement"), "truly," "quite frankly," "to be honest," "let's be clear," "it's worth noting that." - **Vague endorsement**: Cut "worth reading," "worth paying attention to," "worth a look." Say why something matters instead. - **Hedging**: Cut "perhaps," "could potentially," "it's important to note that," "to be clear." - **Compulsive rule of three**: Vary groupings. Use two items, four items, or a full sentence instead of triads. ### Words to replace (Tier 1, always flag) delve / delve into → explore; landscape (metaphor) → field/space/industry; tapestry → describe complexity; realm → area/field; paradigm → model/approach; embark → start/begin; beacon → rewrite entirely; testament to → shows/proves; robust → strong/reliable; comprehensive → thorough/complete; cutting-edge → latest/newest; leverage (verb) → use; pivotal → important/key; underscores → highlights; seamless → smooth/easy; utilize → use; game-changer → describe what changed; unpack → explain/break down; holistic → complete/full; actionable → practical/useful; impactful → effective/significant; learnings → lessons/findings; thought leader → expert; best practices → what works/proven methods; synergy → describe the effect; in order to → to; due to the fact that → because; serves as → is Notable marketing/social words to cut: "nestled," "vibrant," "thriving," "bustling," "showcasing", replace with plain description or specifics. ### Template phrases (avoid) - "a [adjective] step towards [adjective] AI infrastructure" → describe what changed - "Whether you're [X] or [Y]" → pick the actual audience - "I recently had the pleasure of [verb]-ing" → just say what happened - "In today's [X]" / "In an era where" → cut or state specific context - "Here's what's interesting" / "Here's what caught my eye" → let content signal its own importance - "In conclusion" / "In summary" → your conclusion should be obvious - "When it comes to" → talk about the thing directly - "That said" / "That being said" → use "but" or cut ### Structural issues - **Uniform paragraph length**: Vary deliberately. Include 1-2 sentence paragraphs and longer ones. - **Formulaic openings**: Don't lead with broad context ("In the rapidly evolving world of..."). Lead with the news or insight. - **Suspiciously clean grammar**: Fragments, sentences starting with "And"/"But", if natural, keep them. - **Synonym cycling**: Human writers repeat the clearest word. Don't rotate synonyms to avoid repetition. - **Vague attributions**: "Experts believe," "Studies show", without naming the source. Either cite or drop. - **Generic conclusions**: "The future looks bright," "Only time will tell," "One thing is certain", cut. - **Chatbot artifacts**: "I hope this helps!", "Certainly!", "Great question!", "Let me know if you need anything else", remove entirely. - **"Let's" constructions**: "Let's explore," "Let's take a look," "Let's break this down", just start with the point. - **Promotional language**: "nestled in breathtaking foothills," "vibrant hub of innovation", use plain description. - **False ranges**: "from the Big Bang to dark matter", list actual topics or pick the one that matters. - **Title case headings**: Use sentence case. Title case only for the piece's main title, if at all. - **Cutoff disclaimers**: "As of my last update," "I don't have access to real-time data", either find the info or cut. - **Novelty inflation**: "He introduced a term," "a concept nobody's naming", describe what the person actually did with the concept. - **Emotional flatline**: "What surprised me most," "I was fascinated", if the emotion is claimed, the writing should earn it. - **Rhetorical question openers**: "But what does this mean for developers?", if you know the answer, just say it. - **Reasoning chain artifacts**: "Let me think step by step," "Breaking this down," "Step 1:", the reader doesn't need the scaffolding. - **Sycophantic tone**: "Great question!", "Excellent point!", conversational rewards, remove. - **Acknowledgment loops**: Restating the prompt before answering, the reader knows what they asked. - **Excessive structure**: More than 3 headings in under 300 words. 8+ bullets in under 200 words. - **Copula avoidance**: Replacing "is"/"has" with "serves as," "features," "boasts," "presents." Default to "is" unless a more specific verb adds meaning. - **False concession**: "While X is impressive, Y remains a challenge", either make both specific or pick a side. - **Parenthetical hedging**: "(and, increasingly, Z)" / "(or, more precisely, Y)", if it matters, give it its own sentence. - **Confidence calibration phrases**: "Interestingly," "Surprisingly," "Importantly," "Notably," "Significantly", let the fact speak for itself. One "notably" in 2,000 words is fine; three in 500 is AI stacking. - **Over-polishing warning**: Aggressively sanding off every irregularity pushes writing *toward* AI statistical profiles. Natural disfluency and idiosyncratic word choices keep text sounding human. ### When to rewrite from scratch vs, patch If 5+ flagged vocabulary hits across multiple categories, 3+ distinct pattern categories triggered, and uniform sentence/paragraph length: advise a full rewrite. State the core point in one sentence, then rebuild. ### Output format (rewrite mode) Return: 1. Issues found (bulleted list), 2. Rewritten version, 3. What changed (summary), 4. Second-pass audit (check rewrite for remaining AI tells). ### Output format (detect mode) Return: 1. Issues found (grouped by severity P0/P1/P2), 2. Assessment (which flags are clear problems vs, judgment calls).

05-09 · What Is claude.md? Special file

claude.md is a special file Claude reads automatically when it enters your project folder. Think of it as "global instructions", things you want Claude to know every single session, without being asked.

✓ What goes in claude.md

Project-level context (the big picture)
Key conventions and rules
Links to your other context files
Starting instructions for every session

✗ What does NOT go in claude.md

Task-specific details (those go in your prompt)
Sensitive information (no secrets here)
One-time instructions

claude.md is your project's "permanent briefing." Write it once, and Claude consults it on every task.

claude.md

# Claude Instructions ## Always do this - Read all files in ABOUT ME/ before starting any task - Ask me clarifying questions BEFORE executing - Save all outputs to the OUTPUTS/ folder - Use my writing style from about-me.md at all times - When something is unclear, ask, don't guess ## Never do this - Don't start executing without asking questions first - Don't use corporate jargon (see anti-ai.md) - Don't assume, always verify ambiguous instructions - Don't pad responses with filler sentences ## My workspace context [Brief description of what you work on, your team, clients]

05-10 · How Context Files Grow Over Time

Timeline	What to do
Week 1	Start small: about-me.md + claude.md + anti-ai.md. Nothing else needed.
Month 1	Add a rule whenever Claude makes the same mistake twice. Create a template for any recurring task.
Month 3	Remove outdated content. Merge related files. Share templates with your team.

Context files are living documents. The best time to write a new rule is immediately after Claude makes the same mistake twice.

05-11 · Before vs. After: The Context Difference

Scenario: "Write a response to a client who's concerned about timeline delays."

❌ Without context files

"Dear [Client], Thank you for your concern about the timeline. We understand delays can be frustrating. We are working hard to get things back on track."

→ Generic. Vague. Sounds like every other agency email.

✓ With context files (about-me + writing-rules)

"Hi Sarah, Thanks for flagging the timeline concern, it's completely valid and I appreciate you raising it. Here's where we stand: [specific update]. We'll have a revised timeline to you by Friday. Does that work?"

→ Specific. On-brand. Personal. Actionable.

05-12 · Exercise: Write Your First Context File

Wrote about-me.md, ran the interview prompt or wrote manually from the template
Created claude.md, standing instructions Claude follows every session
Created anti-ai.md, your list of writing pet peeves (see Section 6)
All files saved in the ABOUT ME/ folder
Tested with "read my folder" prompt. Claude confirmed it understands

16 Cowork Mode · 10 slides

Claude Becomes a
Real Teammate

Cowork Mode is where Claude transforms from a chatbot into a collaborator. The single biggest productivity jump you'll experience.

06-02 · What Changes in Cowork Mode

Chat mode

Claude can only talk to you
No access to your files
No memory of past conversations
Everything must be typed manually

Cowork mode

Claude sees your entire project folder
Reads any file you point it to
Creates and edits files directly
Works WITH you, not just responding to you

Cowork Mode is where Claude transforms from a chatbot into a collaborator. It's the single biggest productivity jump you'll experience.

06-03 · How Cowork Mode Works

📖

Read

Opens and reads any file in your project folder. Instantly knows your context, drafts, and reference materials, no copy-paste needed.

✍️

Write

Creates new files and edits existing ones. Writes drafts, updates documents, organizes your work directly to disk.

🧠

Think

Reasons through problems step by step. Uses AskUserQuestion to ask YOU clarifying questions before executing.

🔍

Searches through your files for specific information. No more manual Ctrl+F through long documents.

💡

You are always in control. Claude only does what you ask it to. The difference is it now works from your actual files, so it works from facts, not guesses about what you might mean.

06-04 · The "Read Before Write" Principle

A simple rule: before asking Claude to write or change anything, first ask it to read and understand. Prevents hallucinations and wrong assumptions.

Read

"Read this document and tell me what it's about."

Write

"Now update the timeline section based on the new dates."

Verify

"Show me exactly what you changed."

Read first. Write second. Verify third. This three-step rhythm will save you hours of fixing mistakes.

06-05 · What Claude Can (and Cannot) Do in Cowork

✓ Claude CAN

Read and analyze your files
Write and edit documents
Create new files from templates
Search through your project folder
Make suggestions and recommendations
Track changes across versions

✗ Claude CANNOT

Access files outside your selected folder
Remember things from other sessions (without Projects)
Make decisions for you, you're always in charge
Send emails or messages on your behalf
Browse the internet without a web tool
Execute code without your approval

06-06 · First Safe Tasks to Try in Cowork

Read your context files. "Read my about-me.md and tell me what you understand about me."
Create a document. "Create a meeting-notes-template.md in my project folder."
Edit a document. "Add these three bullet points to the project notes."
Ask about your project. "What files do I have in my project folder?"
Draft something. "Draft a status update email using the info in my project notes."
Review your work. "Read this document and suggest concrete improvements."

Start with reading and drafting. Move to editing once you're comfortable. You're always in control.

06-07 · When to Use Chat vs When to Use Cowork

Use CHAT when

Quick questions and brainstorming
Drafting short content (social posts, quick emails)
Learning about a new topic
Getting a second opinion on something

Use COWORK when

Working on an actual document or project
Reading and analyzing your files
Making edits to existing work
Any task that needs your context files

Think of Chat as asking a colleague a question. Cowork is sitting down with them to work on a document together.

06-08 · Cowork: Recurring Example

Let's see how Cowork changes our meeting-to-email scenario:

"Read my meeting-notes.md in the ABOUT ME folder."

→ Claude reads your notes instantly. No copy-paste needed.

"Using my writing-rules.md and about-me.md, draft a meeting summary."

→ Claude combines your context files + notes. Creates a formatted document.

"Draft the follow-up email to Sarah and save it as acme-followup.md."

→ Claude writes the email. Saves it to your OUTPUTS folder. You review and send.

Total time: ~5 minutes. Everything saved to your project folder. Zero copy-paste.

06-09 · Cowork Projects: Persistent Memory

The old Cowork forgot everything between sessions. Cowork Projects fixes this, persistent memory, custom instructions, task history.

💾

Scoped memory

Claude remembers what it did inside this project. "Build on last week's report", it knows what that was.

📋

Custom instructions

Your rules and tone baked in. Every task inside this project follows them automatically.

📚

Task history

Every run stored. Go back and see what Claude did, when, and the exact output produced.

Create your first Cowork Project: Open Claude desktop → Cowork tab → Projects → + → "Use an existing folder" → Select your Claude Cowork folder → Add custom instructions (paste from claude.md)

I just created this Project. Read every file in the folder. Summarize what you know about this workspace: what's here, what I probably use it for, and what instructions you'll follow. If something is unclear, ask me.

06-10 · Cowork Mode Check-In

✏️

Reflect (2 min):
• What felt easy about Cowork Mode?
• What felt unfamiliar?
• What task would you try first in your real work?

Key takeaway: Cowork Mode isn't a different tool, it's the same Claude with more abilities. You already know how to prompt. Now Claude can also read, write, and organize.

06-10 · Your Full Setup. Done ✓

📁

Folder structure

Claude Cowork / ABOUT ME / OUTPUTS / TEMPLATES

📄

claude.md

Standing orders every session follows automatically

👤

about-me.md

Who you are, how you work, your rules for Claude

🚫

anti-ai.md

What makes your work sound generic. Claude avoids it all

🗂️

Cowork Project

Persistent memory, custom instructions, task history

That's it. You've done what most Claude users never do. Claude now knows who you are, how you work, and what you expect, every session.

⚡

Power users: After Cowork Mode, explore the Advanced Track below for Browser, MCP, Skills, File Access, and Scheduled Tasks. Otherwise continue to Building the Habit for daily workflow habits.

22 Building the Habit · 9 slides

From Trying It
to Using It Daily

The habit is simple. The discipline is the hard part. Here's exactly how to build it, and what to do when things go wrong.

07-02 · The Iteration Loop

1. Ask

Write a clear prompt using your framework. Context + Task + Format at minimum.

2. Review

Read Claude's output critically. Actually read it. Don't accept the first draft without reading it.

3. Refine

Tell Claude what to change. Don't rewrite it yourself, tell Claude to fix it.

4. Repeat

2-3 rounds of refinement is normal. First draft is always a starting point, never a finished product.

Good work with AI is iterative. The first response is a starting point, not a finished product. Plan for 2-3 rounds.

07-03 · When Claude Gets It Wrong (And Why)

What went wrong	Why it happened	The fix
Generic, vague output	Not enough context	Add background, audience, specific situation
Wrong assumptions	Missing decision rules	Add a rule to your context files
Outdated information	Claude's knowledge has a cutoff date	Provide current info directly in your prompt
Invented facts (hallucination)	Claude filled gaps with patterns	Always verify facts, names, numbers independently
Misinterpreted task	Instructions were ambiguous	Be more specific; give a concrete example

When Claude gets it wrong, 90% of the time the fix is in the prompt, not the model. What did you leave out?

07-04 · Troubleshooting: Prompt Issues

Problem	Likely cause	Fix
Output is generic	No context	Add background, audience, specific situation
Output too long / too short	No length constraint	Specify word count or "2-3 sentences max"
Wrong tone	No tone guidance	Give examples + describe the voice explicitly
Missing key information	Task too vague	Break into specific sub-tasks
Off-topic tangents	No boundaries set	Add "Don't include X" or "Focus only on Y"

If the output isn't what you wanted, add what's missing to your prompt. One sentence of clarification can change everything.

07-05 · Troubleshooting: Context Issues

Problem	Likely cause	Fix
Claude ignores context files	Files not in project folder	Check files are in ABOUT ME/ folder
Claude forgot earlier instruction	Context window too full	Start fresh conversation with key context
Output doesn't match your voice	No writing rules or anti-ai.md	Create or update your writing rules file
Claude says "they" for your company	Missing rule in context files	Add "Always say 'we', never 'they'" to claude.md
Claude doesn't know your client	No client context	Add client notes to your context files

07-06 · Troubleshooting: Mode & Model Issues

Problem	Likely cause	Fix
Response too slow	Using Opus for simple tasks	Switch to Sonnet for everyday work
Response lacks depth	Using Haiku for complex tasks	Switch to Opus + Extended Thinking
Can't create files	In Chat mode, not Cowork	Switch to Cowork Mode
Claude can't see your files	Wrong folder selected	Check you opened the right project folder
Claude says "I can't do that"	Capability not available in this mode	Check if task needs Cowork mode

When something doesn't work, first check: am I in the right mode? Chat can't write files. Cowork can't browse the web. Use the right tool.

07-07 · When to Start a New Conversation

Start fresh when

Claude keeps repeating itself
You switched to a completely different topic
Conversation is very long (100+ messages)
Claude's responses are getting worse over time

Keep going when

You're still iterating on the same task
Good context is already established
The conversation is still productive

Fresh conversations aren't failure, they're like clearing a whiteboard for a new problem. Smart users start fresh often.

07-08 · Troubleshooting Scenarios

Scenario 1

You asked Claude to "write a proposal" and got back something that sounds like every other agency. What's the most likely fix?

Scenario 2

Claude keeps referring to your company as "they" instead of "we." How do you fix that permanently?

Scenario 3

You're 50 messages into a client project conversation and now need to draft something for a completely different client. What do you do?

💬

Discuss with a partner or write your answers. Then we'll review together.

Answers: (1) Add context: about-me.md + company voice + a real example. (2) Add rule to claude.md and anti-ai.md permanently. (3) Start a completely fresh conversation, different client = different context.

07-09 · Building the Habit: Your 5 Key Habits

Write prompts with context: Role, Context, Task, Format, every single time
Read and verify everything Claude produces before using or sending it
Iterate, first draft is never final, expect and plan for 2-3 rounds
Use Cowork Mode for any project work or file creation
Start fresh conversations when switching tasks or topics

23 Trust & Judgment · 5 slides

Claude is Brilliant.
You're Responsible.

Claude makes you faster. Your judgment makes you good. This distinction is the most important thing in this entire training.

08-02 · Trust, But Always Verify

Claude is fluent, confident, and sometimes wrong. Verification isn't optional, it's the whole game.

❌ Dangerous

"Write a migration script."

→ Runs without review. The script had a destructive operation that wasn't flagged. Disaster.

✓ Smart

"Write a migration script. Before you finish, check: Does it handle rollbacks? Does it preserve existing data? Flag any destructive operations."

→ Reviewed. Caught two issues. Safe.

Claude's output is a first draft from a brilliant but imperfect teammate. Always review before shipping, deploying, or sending.

What to verify: Facts & numbers, factual correctness, security issues, appropriateness of tone, assumptions Claude made.

08-03 · Verification Checklist

Before you use anything Claude produces:

Facts. Are names, dates, numbers, and statistics correct? Claude can and does hallucinate.
Tone. Does this sound like you and your company? Or does it sound like generic AI?
Assumptions. Did Claude assume something about your situation that isn't actually true?
Completeness. Does it cover everything you asked for? Is anything missing?
Accuracy. Is the content actually correct? Or does it merely sound correct?
Safety. Does it contain anything sensitive, speculative, or inappropriate for the audience?

The faster Claude produces output, the more carefully you should review it. Speed doesn't mean accuracy.

08-04 · What NOT to Share With AI

❌ Never share

Passwords, API keys, or credentials
Personal private data (passport numbers, financial details)
Confidential business secrets or anything under NDA
Proprietary code or data without company approval

✓ Generally OK

Meeting notes (without sensitive data)
Draft documents and public information
General project context
Writing samples and templates

⚠️

Rule of thumb: Ask yourself, would I be comfortable sharing this with a contractor? If not, don't share it with AI.

08-05 · Your Judgment Is What Matters

The most important takeaway of this entire training:

Claude is a tool, it doesn't replace your judgment.

You are responsible for everything you produce using AI
You make the final call on what's right and what's ready
You catch the mistakes. You spot the gaps. You decide when it's done.
You choose what to share and what to keep private.

Claude makes you faster. Your judgment makes you good.

AI is the pencil. You're the writer. The pencil doesn't write the book, you do.

24 Closing · 4 slides

You Made It.
Now Start.

You now know more about using AI effectively than most people who use it every day. The rest is practice.

09-01 · What You've Learned

Part	Key takeaway
1 · Foundations	AI is pattern-matching, not magic. Tokens, context windows, the delegation mindset.
2 · The Landscape	Claude is 6 tools. Chat + Cowork cover 95% of daily needs.
3 · Prompting	Role, Context, Task, Constraints, Format, Verify. Context is the most important part.
4 · Your Folder	Claude Cowork / ABOUT ME / OUTPUTS / TEMPLATES. 5 minutes of setup, saves hours.
5 · Context Files	about-me.md + claude.md + anti-ai.md. Write once, use forever.
6 · Cowork Mode	Claude reads, writes, edits files. Read before write. Projects = persistent memory.
7 · The Habit	Ask → Review → Refine → Repeat. Fix the prompt, not the model.
8 · Trust & Judgment	Verify everything. Never share sensitive data. Your judgment is irreplaceable.

09-02 · Your First Real Task

Open Claude in Cowork Mode on your project folder

Pick a real task from today's work

Email, summary, status update, document draft, anything you'd normally spend 20+ minutes on.

Write a prompt using the framework

Role, Context, Task, Constraints, Format.

Review Claude's output critically

Make one round of refinements, then use it

You don't need to master everything today. One good prompt is a win. Do it once, it becomes a habit.

09-03 · Next Steps & Resources

🧪 Practice daily

Use Claude for at least one real task every day for the next two weeks. The habit compounds fast.

📝 Add rules

Every time Claude gets something wrong, add a rule to anti-ai.md or claude.md immediately.

🔗 Connect tools

Settings → Connectors. Add Slack, Google Drive, Notion. Claude reads your actual tools.

👥 Share with team

Share useful prompts and context files. A shared template library benefits everyone.

Resources:

claude.ai. Chat mode for quick tasks
claude.com/download. Desktop app for Cowork Mode
docs.claude.ai. Official documentation and prompt engineering guide
Your team lead. For questions and support

09-04 · Congratulations

You Made It

Congratulations on completing the training.

You now know more about using AI effectively than most people who use it every day. The rest is practice. Every prompt you write, every output you review, every context file you create, that's where the skill builds.

You've got everything you need to start. The only thing left is to begin.

The best time to start using AI well was six months ago. The second best time is right now.

A Appendix · Prompt Patterns

Advanced
Prompt Patterns

Once you're comfortable with the basics, these patterns unlock more precise and powerful outputs.

A-01 · Model Selection Quick Guide

Model	Best for	Speed	Think of it as
Haiku 4.5	Quick tasks, drafts, simple Q&A	Fastest	Your quick assistant
Sonnet 4.6	Everyday work, writing, analysis	Fast	Your main collaborator
Opus 4.6	Complex analysis, strategy, hard problems	Thoughtful	Your senior advisor
+ Extended Thinking	Math, logic, strategy. "show your work" tasks	Slower but better	Available on Opus

Use the lightest model that does the job. Use Extended Thinking when you want Claude to reason through the problem before answering.

A-02 · Extended Thinking: When Claude "Thinks"

Extended Thinking mode makes Claude reason step-by-step before answering. You see the reasoning process. Better for complex problems.

When to use

Complex analysis, strategic decisions, debugging, anything where "show your work" matters.

When to skip

Simple drafts, quick questions, routine tasks, regular mode is faster and equally good.

What you get

Visible reasoning process. Makes hidden assumptions easier to catch. More reliable on hard problems.

A-03 · Prompt Pattern: Role Prompting

Give Claude a specific persona to adopt. Fundamentally changes the lens of every response.

"You are a [role] with expertise in [area]. You communicate in [style]." Examples: - "You are a senior UX writer who specializes in clear, jargon-free error messages." - "You are a financial analyst explaining complex concepts to beginners." - "You are a project manager who writes concise, action-oriented status updates."

💡

Role setting narrows Claude's output space dramatically. A "senior UX writer" produces very different text than a "product manager" given the exact same task.

A-04 · Prompt Pattern: Chain of Thought

Ask Claude to reason step by step before answering. Makes hidden assumptions visible. Easier to catch errors.

"Before you answer, work through this step by step." Examples: - "Before recommending a solution, list the pros and cons of each option." - "First, analyze what's missing from this document. Then suggest what to add." - "Think about what could go wrong with this plan before helping me refine it."

A-05 · Prompt Pattern: Persona + Constraint

Combine role with specific hard limits for precise, bounded output.

"You are a [role]. You MUST [do X]. You must NOT [do Y]." Examples: - "You are a technical writer. You must use plain language. You must NOT use jargon or acronyms without explanation." - "You are a customer support lead. You must acknowledge the customer's frustration. You must NOT promise solutions we can't deliver."

A-06 · Prompt Pattern: Format Constraint

Tell Claude exactly how to structure its response. Eliminates guesswork entirely.

Format instruction examples: - "Respond as a table: Option | Pros | Cons | Recommendation" - "Bullets only. No paragraphs. Each bullet under 15 words." - "Give me 3 options. For each: 2-sentence summary, effort level, risk." - "Write as a memo: TO, FROM, DATE, SUBJECT, then 4 sections."

A-07 · More Context File Tips

Keep files focused. One topic per file. Don't dump everything into about-me.md.
Use plain markdown (.md). Claude reads it cleanly. No Word docs in the context folder.
Set a quarterly reminder. Review and update context files every 3 months.
Add real examples. A "good email / bad email" example is worth more than any description.
Share with your team. A shared template library benefits everyone and compounds over time.
Trim when it gets long. If about-me.md exceeds 2,000 words, ask Claude to help you condense it.

A-08 · Building a Template Library

Find a task you do at least weekly

Meeting recap, client email, status update, project brief, weekly report.

Write the best prompt you can for it

Use all 6 parts of the framework.

Replace specific content with [PLACEHOLDERS]

Remove names, dates, and details.

Save to TEMPLATES/ folder

Name it clearly: "meeting-recap.md", "client-email.md"

# Template: meeting-recap.md Task: Summarize this meeting. Format: Key decisions → Action items (with owner) → Open questions Tone: Brief, factual, no filler Include: Who attended, what was decided, what happens next Context: [PASTE MEETING NOTES HERE]

A-09 · Bonus: What Are Skills?

Skills are packaged prompt workflows that Claude can run automatically, like "super templates" for complex recurring tasks.

What it is

A SKILL.md file in a folder that defines what Claude does, when to use it, and the step-by-step process.

When to use

Once you're comfortable with prompting and Cowork. Skills are for tasks you run 2-3x per week with a consistent structure.

Examples

Meeting recap generation, client email drafting, project status updates, research summarization.

💡

Skills are more advanced. Get comfortable with basic Cowork first. Come back to Skills once you've built the habit.

A-10 · What Makes a Good Skill

Specific. Does one thing well, not everything vaguely
Reusable. You use it at least once a week
Structured. Clear steps, not a wall of text
Tested. You've run it 3+ times and refined it
Shared. Your team can use it too

B Appendix · Practice & Quick Reference

Practice & Quick Reference

Extra exercises, cheat sheets, and everything you need in one place.

A-12 · More Troubleshooting Tips

Symptom	Check	Fix
Claude is repetitive	Context window too full	Start fresh conversation
Claude ignores instructions	Prompt is too long	Put key instructions first and last (Claude pays most attention to beginning and end)
Claude is too formal	No tone guidance	Add "Write like a human, not a corporation" to your prompt or anti-ai.md
Claude says "I don't have enough info"	Missing context	Add more specific details about the situation
Output is too long	No length constraint	Add "Keep this under [X words/sentences]"

A-13 · Extra Practice: Meeting Recap

Exercise

You attended a 30-minute project check-in. Key points: Budget approved for Q3 campaign. Timeline moved up 2 weeks (deadline: June 15). Client requested weekly check-ins. Two new stakeholders added: Sarah (Finance) and Mike (Creative).

✏️

Your task: Write a prompt that produces a complete meeting recap with: summary, key decisions, action items, and next meeting date. Write the prompt before reading below.

Example answer: Role: You're a project coordinator attending a Q3 campaign check-in. Context: [paste notes]. Task: Write a full meeting recap. Format: Summary (3 bullets) → Decisions (numbered list with owners) → Action items (due dates) → Next steps. Constraints: Under 300 words. Flag any action items that don't have a clear owner.

A-14 · Extra Practice: Client Communication

Exercise

A client emailed asking why a deliverable is late. The delay is due to a third-party vendor who provided incomplete assets. The relationship is 2 years old. The client has been patient but this is the second delay this quarter.

✏️

Your prompt should produce a response that:
• Acknowledges the delay honestly
• Explains the cause without sounding like an excuse
• Provides a new timeline with some buffer
• Rebuilds confidence
• Uses your company's professional-but-warm voice (reference your writing-rules.md)

A-15 · Extra Practice: Research & Summary

Exercise

You have a 20-page industry report. You need to: (1) extract the 5 most important findings, (2) explain each in 1-2 sentences, (3) note implications for your company, (4) suggest 3 potential actions.

✏️

Write a single prompt that accomplishes all 4 goals. Hint: Include the format you want, how long each section should be, and a verification step to flag anything uncertain or that requires further research.

A-16 · Resources & Links

🌐 claude.ai

Chat mode, quick tasks, brainstorming, summaries. Free plan available.

💻 claude.com/download

Desktop app. Cowork Mode, Projects, file work.

💰 claude.com/pricing

Free, Pro ($20/mo), Max ($100-200/mo).

📖 docs.claude.ai

Official documentation and prompt engineering guide.

💡

Practice plan: Use Claude for at least one real task every day for the next 14 days. By day 14 it will feel as natural as email. Try rewriting one weak prompt per day. Share one useful prompt with a teammate each week.

A-17 · Common Terms Reference

Term	What it means
AI	Computer systems that can perform tasks requiring human-like intelligence
LLM	A model trained on vast text data to understand and generate language
Token	~¾ of a word, the unit AI reads and writes in
Context Window	How much text AI can process at once (Claude: ~200K tokens)
Prompt	Your instruction to the AI
Output	What the AI returns in response
Hallucination	AI confidently stating something incorrect
Sycophancy	AI tendency to agree even when the user is wrong
Model	A specific AI version (Haiku, Sonnet, Opus)
Fine-tuning	Customizing AI for a specific use case
Cowork Mode	Claude desktop app that reads/writes files on your computer
Extended Thinking	Mode where Claude reasons step by step before answering
AskUserQuestion	Cowork feature where Claude asks you questions before executing a task

A-18 · Quick Reference Card

prompt-framework.md, bookmark this

# The 6-Part Prompt Framework Role → "You are a [role] who [expertise]" Context → "Here's what you need to know: [background, audience, situation]" Task → "Do this: [specific instruction]" Constraints → "Stay within: [length, tone, avoid X]" Format → "Structure your response like: [format description]" Verify → "Before finishing, check: [verification criteria]" ───────────────────────────────────────── Always: Read first. Ask second. Verify third. If stuck: Add context. If repeated mistake: Start fresh + update context files. If wrong: Be more specific. Give an example. If too generic: Add your context files to the prompt.

A-19 · Prompt Cheat Sheet by Task

Task	Template
Summarize	"Summarize this in [X] bullet points. Include key decisions, open questions, and action items."
Draft	"Draft a [document type] for [audience]. The goal is [outcome]. Use a [tone] tone."
Edit	"Review this for [clarity/tone/completeness]. Suggest specific changes. Explain why."
Brainstorm	"Generate [X] ideas for [goal]. For each: one-sentence summary, effort level, potential impact."
Compare	"Compare [A] and [B] across [criteria]. Present as a table with a recommendation."
Research	"Summarize the key arguments for and against [topic]. Flag anything uncertain."
Status update	"Write a status update for [project]: progress this week, blockers, next steps. Under 200 words."

A-20 · What's Next for AI

The AI landscape is moving fast. Here's what's coming:

Near-term (6-12 months)

Better integration with tools you already use
AI agents that run multi-step workflows
Improved accuracy, fewer hallucinations
More specialized models for specific industries

What stays the same

Your judgment is still what matters most
Context and clarity will always improve output
Verification will always be necessary
The human in the loop is irreplaceable

The tools will evolve. The skill of working with AI, context, clarity, judgment, will only become more valuable.

25 TL;DR · Takeaways

TL;DR: Just Tell Me What to Do

1 Create your project folder

5 minutes. Claude Cowork/ with ABOUT ME/, OUTPUTS/, TEMPLATES/. This is the foundation. Without it, every session starts from zero.

→ Do this first, before anything else

2 Write about-me.md + claude.md

Use the interview prompt in this training. Tell Claude who you are, how you work, and what you expect. Write once, use forever. This is what makes output sound like you.

→ Create both files this week

3 Use the 6-part prompt

Role · Context · Task · Constraints · Format · Verify. At minimum: Context + Task + Format. The 30 seconds you spend setting context saves 30 minutes of editing.

→ Bookmark the prompt framework (section A-18)

4 Verify before you ship

Two-pass reading: facts first, tone second. Assume Claude is confidently wrong until you prove otherwise. Never send AI output without review. Ever.

→ Use the verification checklist (section 8-03)

5 Iterate 2-3 rounds

First draft is never final. Tell Claude what to fix - don't rewrite it yourself. 2-3 rounds of refinement is normal, not failure. Quality is in the iteration.

→ Expect refinement, budget for it

6 Start fresh when context shifts

New task = new conversation. Different client = different conversation. When responses get worse, you've filled the context window. Start fresh and bring key context forward.

→ Fresh conversations are not failure

The One Rule That Trumps Everything

Claude drafts. You decide.

AI removes the friction of getting words on the page. It does not remove the responsibility of deciding whether those words are right. Your judgment is the bottleneck - not the speed of generation.

❌ Low-value use

One-shot prompts, no context, accept first output
Copy/paste between tools
Treat Claude like a search engine

✓ High-value use

Rich context + verification + iteration
Work in Cowork Mode with project folder
Treat Claude like a collaborator

⏱️

Week 1

Create folder + context files. One real task per day in Cowork.

📈

Week 2

Build 2-3 prompt templates in TEMPLATES/. Add rules as mistakes happen.

🔁

Month 1

Review context files. Remove stale rules. Share a template with a teammate.

🧠

Ongoing

Every time Claude gets it wrong, fix the prompt or add a rule. Compound.

You now know more about using AI effectively than most daily users. The gap between knowing and doing is one folder, one file, one prompt. Start today.

17 Browser · Web Search & Fetch

Browse the Web from Your Terminal

Claude can search the web, fetch documentation, and read live pages - all without leaving your editor. No more context-switching to a browser tab.

⚡

Advanced Track: This section is for developers and power users. Feel free to skip and return later - the main track continues at Building the Habit.

06b-01 · Web Search

Claude can search the web for current information, documentation, code examples, and more. The results are brought directly into your conversation context.

🔍 Live Search

Ask Claude to search the web: "Find the latest React 19 release notes." Claude returns summarized results with source links.

📚 Documentation Lookup

"Search for the Prisma migration API reference." Claude finds the relevant docs and incorporates them into your working context.

🧠 Research Assistant

Combine search with your project context: "Search for best practices in Next.js 15 server components, then review my page.tsx against them."

💡

Prompt pattern: Be specific about what you need. "Search for X" gets links. "Search for X and summarize the key points" gets analysis. "Search for X and apply it to my code in Y" gets action.

06b-02 · Fetch & Browse Pages

Beyond search, Claude can read the full content of any public URL. This is useful for API documentation, blog posts, error pages, or any live web content.

✓ What you can fetch

API documentation pages for specific library versions
GitHub repos, issues, and pull requests
Technical blog posts and tutorials
Package registry pages (npm, PyPI, crates.io)
Your own deployed apps for debugging

⚠️ Limitations

Requires a public URL (no authenticated/internal pages)
Some JavaScript-rendered pages may not load fully
Large pages are summarized, not returned verbatim
Rate limits may apply for aggressive fetching

Fetching a URL is like asking Claude to "read" a page for you. It processes the content and tells you what's relevant - you don't need to scan the page yourself.

06b-03 · When to Browse vs. Search

Knowing which tool to use saves time. Search is for discovery; fetch is for depth. Use search when you don't know the exact source, and fetch when you have a specific page in mind.

🔎 Use Search When

You need to find something: "Search for recent Tailwind CSS v4 changes." You don't know the exact URL or source. Let Claude find it.

📄 Use Fetch When

You have a specific URL: "Fetch https://example.com/docs/api and tell me how to use the createClient function." You want deep analysis of one page.

🔄 Combine Both

Search first to find the right page, then fetch it: "Search for the Prisma migration guide - then fetch the top result and help me write my schema."

⚠️

Remember: Web search and fetch count toward your API usage. Use them intentionally - don't make Claude re-search for the same page multiple times in one session.

18 Connections · MCP Protocol

Connect Claude to Your Tools

MCP (Model Context Protocol) is how Claude connects to your database, APIs, file storage, and internal tools. It turns Claude from a standalone chat into an integrated part of your development ecosystem.

06c-01 · What is MCP?

MCP is an open standard that lets AI models interact with external systems through a unified protocol. Think of it as "USB-C for AI" - one standard way to connect any tool to any AI model.

🔌 Standardized Interface

Every MCP server exposes the same interface: tools, resources, and prompts. Claude discovers what's available when the connection starts - no manual configuration per tool.

🔒 User-Controlled Access

Claude asks for permission before using any connected tool. You approve each action individually. Connections are not backdoors - they're gates you control.

🌍 Open Ecosystem

MCP servers exist for databases (Postgres, SQLite), cloud storage (Google Drive, Dropbox), project management (Jira, Linear), communication (Slack), and more. Anyone can build one.

MCP is not a Claude-specific feature - it's an industry protocol. Connecting your tools through MCP means you're using an open standard that other AI tools can also adopt.

06c-02 · Setting Up Connections

Connections are configured per-project in your Claude settings. You define the MCP server endpoint, and Claude handles the rest.

✓ Step-by-step

Find or install an MCP server for your tool (npm packages, Docker containers, or custom scripts)
Add the server configuration to your project's CLAUDE.md or settings
Claude discovers available tools when the session starts
Use natural language: "Query the database for recent orders" or "Create a Jira ticket"

⚠️ Example Config

Database: MCP server connects to Postgres, exposes read-only query tools
File storage: MCP server connects to Google Drive, provides file search and read tools
Project management: MCP server wraps the Jira API, exposes create/update/search tools
Each server defines its own tool set - Claude adapts automatically

💡

Start small: Pick one MCP server (your database is a great first choice) and get it working. Add more connections as you find real use cases. Don't try to wire everything at once.

06c-03 · Real-World Use Cases

Connectors bridge Claude to specific services. Here are two practical examples of how you'd use them in daily work.

✈️ Booking.com Connector

Find the cheapest flight from New York to Tokyo, no red-eye flights. The connector queries Booking.com's inventory and filters by your preferences - price, duration, departure time. Claude presents the top options in a clean summary without you opening a single browser tab.

📧 Gmail Connector

Unsubscribe me from all promotional emails I received this week. The connector scans your inbox, identifies marketing and spam messages, and processes unsubscribe requests - or drafts them for your review before sending.

The real power of connectors is not that Claude can use tools - it's that you can describe what you want in natural language, and Claude figures out which tool to use, with what parameters, in what sequence.

06c-04 · Security & Best Practices

Connections are powerful - and with power comes responsibility. Claude has built-in safeguards, but you should understand how they work.

🛡️ Permission Prompts

Every MCP tool call requires your approval. Claude shows you exactly what it's about to do: which tool, what parameters. You approve or deny each time.

🔐 Read-Only by Default

Configure MCP servers with the minimum access needed. Use read-only database users, scoped API tokens, and sandboxed environments where possible.

📋 Audit Trail

All MCP interactions are logged in your session history. You can review what Claude accessed and when. This makes troubleshooting and security review straightforward.

The best security practice with MCP: configure each connection with the least privilege it needs, and review the permission prompt before approving any action. Treat each connection like a new tool - learn what it can do before relying on it.

19 Skills · Expert Assistants On Demand

Skills: Expertise on Tap

Skills are pre-configured expert capabilities you invoke with /skill commands. They bundle specialized prompts, tools, and context into one command - like having a specialist join your conversation on demand.

06d-01 · Built-in Skills

Claude ships with a growing library of built-in skills for common development tasks. Each one is optimized for a specific job.

🔍 Code Review

/code-review analyzes recent changes for quality, security, and maintainability. It checks for common issues and generates structured reports with severity levels.

🛡️ Security Audit

/security-review scans for vulnerabilities: hardcoded secrets, injection risks, unsafe patterns. Flags issues by severity with remediation steps.

📖 Docs Lookup

/docs-lookup fetches library documentation on demand. "How do I use Prisma transactions?" - it searches docs and returns relevant examples.

⚡ Build Fixer

/build-error-resolver diagnoses build failures. Paste your error message and it walks through the stack trace, identifies root cause, and suggests fixes.

🧪 TDD Guide

/tdd-guide enforces test-driven development: write the test first, implement, verify. Keeps you honest about test coverage.

🔧 Refactor Cleaner

/refactor-cleaner identifies dead code, unused imports, and opportunities to simplify. "Clean up this module" - it analyzes and removes what's not needed.

Skills are not just prompts - they're bundled capabilities. A skill can invoke tools, search your codebase, run analysis, and return structured results. One command, multiple operations.

06d-02 · Creating Custom Skills

Teams and individuals can create their own skills for recurring workflows. This is where skills become force multipliers - you encode your team's best practices once and share them.

✓ What makes a good custom skill

A specific, repeatable task (deployment checklist, onboarding review)
Clear success criteria (what "done" looks like)
Access to the tools it needs (database, linting, etc.)
A well-defined scope (does one thing well)

⚠️ Writing your first skill

Start with a prompt: what would you tell a junior dev to do?
Add tool access: what should the skill be able to run?
Define output: what does the skill return? A report? A fix? An analysis?
Test iteratively: run the skill, refine, repeat

💡

Start with one: Identify your most common recurring task - the thing you explain to Claude multiple times per week. Turn that into a skill. One well-built skill saves hours per week.

06d-03 · When to Use Skills

Skills are best for tasks that follow a repeatable pattern. They're not meant for open-ended exploration - that's what regular conversation is for.

✅ Good for Skills

Running a code review before PR submission. Auditing for security issues. Checking test coverage. Formatting code to team standards. Diagnosing build errors.

❌ Not for Skills

Brainstorming architecture. Debugging a novel issue. Learning a new concept. Exploratory coding. Skills automate - they don't explore.

🔄 Skill + Conversation

Use a skill as the starting point: run /code-review to check changes, then discuss findings in normal conversation. Let skills handle the structured work while you focus on judgment calls.

Think of skills as macros for expertise. They don't replace thinking - they eliminate repetitive setup so you can focus on the parts that need your judgment.

20 File Access · Read, Write, Search

File Access: Your Project, Claude's Workspace

Claude operates directly on your file system. It reads, creates, and edits files within your project - always with your awareness and control. This is how Claude goes from conversation partner to active contributor.

06e-01 · Reading & Searching

Claude can read any file in your project and search across your entire codebase. This is how it understands your code before making changes.

📖 Read Files

Claude reads files to understand your code: "Show me the database schema file." It reads the content and can explain, analyze, or suggest changes to it.

🔎 Search Across Files

"Find all places where we call the payment API." Claude greps your codebase, finds every reference, and presents them in context - no manual searching.

📋 Multi-File Analysis

"How does the auth flow work?" Claude reads the relevant files (routes, middleware, handlers) and traces the full flow end-to-end, showing you how pieces connect.

Claude does not have a "memory" of your entire project at once. It reads files you ask about or that are relevant to your request. Be specific about what you want it to look at, or ask it to explore first.

06e-02 · Creating & Editing Files

This is where the real work happens. Claude can create new files, edit existing ones, and restructure code - all with your explicit approval.

✓ Edit workflow

You ask for a change: "Add input validation to the signup form"
Claude reads the relevant files to understand current code
Claude proposes changes as a diff - you see exactly what changed
You approve or request adjustments before the edit is applied

⚠️ What editing means

Claude suggests changes; you control what gets applied
Every edit shows a diff before application - review it
Claude can create new files, modify existing ones, or delete unused code
Edits are atomic: you approve or reject each one individually

⚠️

Golden rule: Always review the diff before approving. Claude is highly capable but can make mistakes. The diff is your safety net - use it every time.

06e-03 · Real-World Example: Tax Receipts

Here's a complete file access workflow that combines reading, creating, and analysis - all in one natural language request.

✓ The Request

"I need to file my taxes. I have receipt photos on my desktop. Create a folder called Tax-2025, dump all the receipt images into it, then output an Excel spreadsheet with a full breakdown of each expense including date, vendor, amount, and category."

⚠️ What Claude Does

Reads the image files from your desktop (using OCR to extract text from receipt photos)
Creates the Tax-2025 folder and organizes all receipts inside
Extracts structured data: dates, vendor names, totals, line items
Generates an Excel (.xlsx) file with columns for date, vendor, amount, category, and file name
Presents a summary of total expenses by category for your review

This example shows the full file access loop: read from one location, process the data, create new files, output organized results. One request, multiple file operations, zero manual work.

06e-04 · Permission Model

Claude has a layered permission system that lets you control exactly what it can access. Understanding this system helps you work confidently.

🔒 Project Scope

Claude only operates within your selected project folder. It cannot access files outside this boundary without explicit permission. Your system files are off-limits.

👁️ Read vs. Write

Reading files is typically automatic. Writing requires approval. You can configure Claude to ask for permission on reads too, or to auto-approve certain write patterns.

🚫 Sensitive Files

Files like .env, credentials, and private keys are never read without explicit request. Claude knows to avoid accessing secrets unless you specifically ask it to.

Permission mode is your comfort level dial. Start with explicit approval for everything. As you build trust, you can relax it for routine operations - but the safety net is always there if you need it.

21 Scheduled Tasks · Automate & Monitor

Scheduled Tasks: Let Claude Watch for You

Claude Code can run tasks on a schedule - reminders, recurring checks, and automated monitoring. Set up once, and Claude keeps you informed without constant manual checking.

06f-01 · One-Shot Reminders

The simplest scheduled task: a single notification at a specific time. Use these for time-sensitive callbacks during your workday.

⏰ Time-Based Reminders

"Remind me to check the deploy at 2pm." Claude schedules a single notification. When the time comes, you get a message in your active session - no separate calendar or alarm needed.

⏱️ Duration Reminders

"Remind me in 30 minutes to review the PR comments." Claude sets a timer relative to now. Useful for not getting lost in a task while something else is cooking.

🔄 Conditional Reminders

"Tell me when the build finishes." Claude monitors a process and notifies you when it completes. You can switch to other work without polling for results.

💡

Use case: One-shot reminders excel during multi-tasking. While Claude works on a long-running task, set a reminder to check back - you stay productive without context-switching every few minutes.

06f-02 · Recurring Jobs

For ongoing monitoring and regular checks, set up recurring tasks that fire on a schedule. These run automatically until you cancel them.

✓ What to schedule

Health checks: "Check staging server every morning at 9am"
Standup prep: "At 9:15am, summarize yesterday's changes"
Deploy monitoring: "Every hour during deployment, check if the rollout progressed"
Cleanup: "Every Friday, list stale branches that can be deleted"

⚠️ Job lifetime

Recurring jobs auto-expire after 7 days to prevent abandoned tasks
Jobs are session-bound - they stop when your Claude session ends
You can cancel any job at any time with a simple command
Jobs survive across context-window resets within a session

Recurring jobs are best for daily operational tasks - things you'd otherwise put on a calendar reminder but that require Claude's context to execute properly. The 7-day expiry is a safety net, not a limitation.

06f-03 · Task Management During Sessions

You can view, cancel, and manage all your active tasks at any time. This gives you full visibility into what's running and when.

📋 List Active Tasks

Ask "What tasks are running?" to see all scheduled jobs, their next fire time, and their status. Clean up tasks you no longer need.

❌ Cancel Tasks

"Cancel the health check reminder" stops a specific task. "Cancel all tasks" clears the entire schedule. Tasks are removed immediately.

🛠️ Modify on the Fly

"Change the health check to every 2 hours instead of 1." Cancel and recreate with the updated schedule. Tasks are cheap - don't hesitate to adjust.

⚠️

Session scope: Scheduled tasks are tied to your active Claude session. If you close the session, the tasks stop. For persistent automation, use system-level scheduling (cron, CI/CD) - Claude tasks are for development-time assistance, not production infrastructure.

C Appendix · PDF → MD Skill

PDF to Markdown: Save Tokens, Read Better

Claude reads Markdown efficiently. PDFs - especially scanned ones - consume excessive tokens and often produce garbled output. Converting PDFs to Markdown before uploading reduces token usage by 50-80% and dramatically improves output quality.

The Problem: PDFs Eat Your Token Budget

When you batch-upload PDFs to Claude, you'll notice your token budget exhausted very quickly. That's because Claude must process raw PDF data - layout information, fonts, sometimes embedded images - all of which count toward your token limit. A 20-page PDF that uses 40,000 tokens in its original format might use only 8,000 tokens when converted to Markdown.

The fix is simple: Convert PDFs to Markdown before uploading. Markdown is pure text with lightweight formatting - Claude reads it directly, and you get 3-5x more content per token.

C-01 · The Skill: What It Does

This skill reads PDF files through Cowork Mode, extracts their content, and saves clean Markdown versions. After the first conversion, you work with the Markdown files - the PDF is processed once and the tokens are spent once.

📥 Step 1

Place your PDFs in a Cowork-accessible folder (e.g., ~/Documents/pdf-input/).

⚡ Step 2

Run the skill: /pdf-to-md. Claude reads each PDF and converts it to clean Markdown.

📤 Step 3

Claude saves the .md files side by side with the originals. Upload those to any future conversation.

💡

Why this works: The token cost of reading the PDF happens once. Every conversation after that uses the lean Markdown files. Over time this saves far more tokens than the one-time conversion cost.

C-02 · Install the Skill

Create a file called pdf-to-md.md in your project's skills folder (e.g., ~/.claude/skills/pdf-to-md.md or ./skills/pdf-to-md.md). Paste this content:

--- name: pdf-to-md description: Convert PDF files to Markdown to save tokens and improve readability --- # PDF to Markdown Convert PDF files in a folder to clean Markdown format. ## When to use You have PDFs you want to work with in Claude. Instead of uploading them raw (which wastes tokens on layout data), convert them to Markdown first. ## Steps 1. Ask the user which folder contains PDFs. Default: working directory. 2. Find all `.pdf` files in that folder. 3. For each PDF: a. Read the file through Cowork Mode's file access. b. Extract the text content preserving headings, lists, tables, and structure. c. Format everything as clean Markdown (no extraneous layout data). d. Save the output as `[filename].md` next to the original PDF. 4. Report which files were converted and where the output was saved. 5. Remind the user to use the `.md` files instead of the PDFs in future conversations. ## Notes - This works best on text-based PDFs. Scanned documents will have limited extraction quality. - Tables may lose some formatting in conversion - review and clean up before using. - The skill processes each PDF once. The resulting Markdown files can be reused indefinitely.

C-03 · How to Use It

Once installed, invoke the skill anytime:

In Cowork Mode

Just type /pdf-to-md in your conversation. Claude will ask where your PDFs are, read them, and save clean Markdown versions.

In Claude Code

Same command: /pdf-to-md. Claude Code can also run marker-pdf for better accuracy on complex documents (see C-04).

The skill does the work once. From that point on, you drag in the Markdown files - faster uploads, fewer tokens, better results.

What AI Actually Is

01. The Core Idea

02. The Three Ingredients

03. What This Means for You

04. What AI Is NOT

How AI Changed (And Why It Matters)

01. The Three Phases of Modern AI

Phase 1: Rules + Search (Pre-2022)

Phase 2: The Breakthrough (Late 2022)

Phase 3: From Chat to Work (2024-2025)

02. Old Mental Model vs. New

03. Why This Confuses People

Strengths & Limits

01. What AI Excels At

02. What AI Is Bad At

03. The Critical Distinction

04. The Competency Map

How to Think AboutNew Tools

01. The Fundamental Question

02. The Evaluation Framework

Does it change the interface?

Does it access new data?

Does it change who verifies?

03. What Actually Changes vs. What Does Not

04. The Trap of "Just Wait for the Next One"

The Human Role

01. You Are the Editor

02. What Only You Can Do

03. The Taste Gap

04. The Trust Paradox

How to ReadAI Output

01. The Fluency Problem

02. The Two-Pass Reading Method

Pass 1: Substance Scan

Pass 2: Fit Check

03. Common Failure Modes

04. The Verification Checklist

AI Safety & Privacy

01. The Core Rule: What You Put In Matters

02. Data Handling by Platform

03. Privacy Best Practices

04. The Hallucination Risk (Revisited)

05. The Trust Boundary

Know your company's policy

Check before sharing client data

Use enterprise tiers for sensitive work

When in doubt, leave it out

The CollaborationPattern

01. The Core Loop

Orient

Direct

Review

Iterate

02. Surface-Level vs. Deep Collaboration

03. The Feedback Loop

04. Common Anti-Patterns

05. The Meta-Skill

06. Durable Skills Summary

Why Set Up Claude?

08-01 · The Real Cost of Not Setting Up

08-02 · AI as Your Second Brain

Externalize memory

Think in public

Process at scale

Build a knowledge base

Iterate faster

08-03 · What Setup Unlocks

08-04 · What Setup Is (And Isn't)

The Way You WorkIs About to Change

01-01 · Old Paradigm vs. New Reality

01-02 · Tokens: How AI Reads Your Words

01-03 · The Context Window: How AI Remembers

01-04 · The Beginner Trap: Why Conversations Break

01-05 · The 4-Step Delegation Flow

Give context

Set expectations

Define the task

Iterate & verify

01-06 · Exercise: Reflect on Your Current Use

Meet YourNew Teammate

How to Think About
New Tools

How to Read
AI Output

The Collaboration
Pattern

The Way You Work
Is About to Change

Meet Your
New Teammate

How to Talk to Claude
So It Actually Works

The Files That Make
Claude Actually Yours

Claude Becomes a
Real Teammate

From Trying It
to Using It Daily