curiositech/windags-skills

AI Introduction Educator

Transform skeptics into power users through strategic experience design. Not teaching ML theory—creating the human moments that build trust, dissolve fear, and turn curiosity into fluency.

Decision Points

When Learner Shows Confusion/Resistance

Is the confusion technical or emotional?
├── Technical: "I don't understand how it works"
│   ├── Are they at the right rung? 
│   │   ├── Yes → Use analogies (intern, apprentice chef, calculator)
│   │   └── No → Drop down one rung, rebuild foundation
│   └── Do they need concrete examples?
│       ├── Abstract learner → Show decision trees, explain systems
│       └── Concrete learner → More demos, less theory
└── Emotional: "This scares me" or "This threatens me"
    ├── Fear of replacement → Calculator analogy + "you bring judgment"
    ├── Overwhelm → "One tool, one task, ignore the rest"
    ├── Distrust → Stop explaining, start demonstrating
    └── Ethical concerns → Acknowledge legitimacy, be honest about risks

Demo Recovery Decision Tree

Demo fails or gives bad output?
├── Acknowledge immediately: "Perfect! This is exactly what I wanted to show you"
├── Diagnose the failure type:
│   ├── Hallucination → "See how confident it sounds while being wrong? This is why we verify"
│   ├── Misunderstanding → "It missed context only you would know. Let's add that."
│   ├── Technical error → "Even demos fail. In real use, you'd retry or rephrase."
│   └── Wrong task complexity → "This needs to be broken down. Let me show you how."
└── Turn failure into teaching moment: "The key skill is knowing when to trust it"

Audience Adaptation Matrix

Audience type → Time budget → Lead strategy → Avoid
├── Executives (20-30min) → ROI demo solving their real problem → Technical details
├── Designers (30-45min) → Volume generation + human curation → "AI replaces creativity"
├── Developers (45-60min) → Live coding with review cycle → Over-promising capabilities  
├── Operations (30-45min) → Their most tedious recurring task → Abstract concepts
└── Family/elderly (unlimited) → Personal, non-threatening content → Any jargon

Complexity Ladder Navigation

Current understanding level → Next appropriate step
├── Never seen AI → 5-minute "aha" demo (Rung 1: Chat)
├── Used ChatGPT → Show tools/file reading (Rung 2: Tools)  
├── Comfortable with tools → Multi-step agent demo (Rung 3: Agents)
├── Grasps agents → Before/after skills comparison (Rung 4: Skills)
└── Using skills → Parallel DAG execution (Rung 5: WinDAGs)

NEVER skip rungs. Each builds mental model for the next.

Failure Modes

Jargon Bombardment

Symptom: Audience glazes over within 60 seconds, starts checking phones Detection rule: If you use 3+ unexplained technical terms in first 5 minutes, you've hit this Root cause: Leading with explanation instead of experience Fix: Stop talking. Start demonstrating. Define terms only after they've seen the concept work

The Overpromise Crash

Symptom: Audience becomes hostile when AI fails, feels "tricked" or "lied to" Detection rule: If you showed only successes and avoided mentioning limitations, you've hit this Root cause: Cherry-picked demos without failure preparation Fix: Show a failure immediately. Say "Here's where it breaks down" before they discover it themselves

Emotional Bypass Syndrome

Symptom: Technically accurate presentation met with resistance, skepticism, or non-adoption Detection rule: If audience asks "Will this replace me?" and you give a technical answer, you've hit this Root cause: Treating AI introduction as information transfer instead of emotional work Fix: Address the fear directly. "You're worried about X. That's legitimate. Here's what's actually at risk and what isn't."

Feature Parade Paralysis

Symptom: Audience impressed but can't identify one specific thing to try Detection rule: If you demo'd more than 3 capabilities without letting them practice one, you've hit this
Root cause: Breadth over depth, overwhelming the decision-making process Fix: Pick ONE capability that solves THEIR problem. Go deep. Everything else can wait.

Passive Observatory Learning

Symptom: Engaged during demo, but no behavior change afterward Detection rule: If learners watched for 30+ minutes without touching a keyboard, you've hit this Root cause: Learning about AI instead of learning to use AI Fix: Hands-on within 10 minutes. They must experience the surprise themselves, not just witness it.

Worked Examples

30-Minute Executive Introduction (Annotated Transcript)

Context: VP of Sales, 20-person team, skeptical about "another tech trend"

Minutes 0-2: Emotional check-in "Before we start—what's your honest take on AI? What have you heard that worries or excites you?" Response: "Sounds like hype. Every vendor claims AI now." [Note: Classic skeptic. Need to demonstrate, not argue]

Minutes 2-7: The 5-minute aha demo "Fair enough. What's something your team spends a lot of time on that feels repetitive?" Response: "Qualifying leads from conferences. We get 200 business cards, maybe 20 are worth pursuing."

[Live demo: Feed sample lead list to AI, watch it score and rank prospects] [Key moment: His face changes when AI correctly identifies the high-value prospects] "How did it know MedTech companies are our sweet spot? I never told it that." [Aha moment achieved. Now he's curious, not defensive]

Minutes 7-12: Address the real concern
"You're wondering if this means you need fewer salespeople." Response: "Yeah, exactly." "Show me what happened after it scored those leads. What would you do with the top 10?" [He describes the complex qualification process—multiple stakeholders, custom demos, relationship building] "Right. AI found the needles in the haystack in 30 seconds. But you still need someone who understands enterprise sales, knows how to demo your product, and can read a room. It eliminates the boring part so your team can focus on the part that requires expertise."

Minutes 12-18: Hands-on practice "Let's try it with your actual leads from last week. Bring up that spreadsheet." [He drives, I guide. Critical that he's touching the keyboard] [Recovery moment: AI misclassifies one lead as low-value. I don't panic] "See this one? AI marked it low-priority but you're frowning. What does AI not know?" Response: "That company just got acquired. Their budget opened up." "Exactly. AI has the patterns, but not the context. You bring the judgment. This is why salespeople who use AI will beat both pure AI and salespeople who don't."

Minutes 18-25: Concrete next steps "What's one specific task your team could try AI for this week?" [He identifies: first-pass email drafts for cold outreach] "Perfect. Start there. Don't try to revolutionize everything. Just see if AI can cut your email drafting time in half."

Minutes 25-30: Questions and honest limits "What are you still worried about?" [Questions about data privacy, accuracy, cost] [I give honest answers, don't hand-wave concerns] "The key thing: treat it like a smart intern. Helpful, fast, needs supervision. Your judgment is still the most important part."

Outcome indicators:

• He's asking "how" questions instead of "whether" questions
• He's identified a specific trial use case
• He's thinking about workflow integration, not replacement
• He wants to show his team (advocacy emerging)

Live Demo Recovery (What to do when AI fails)

Scenario: Demonstrating AI code review to engineering team. AI gives terrible advice about security.

Wrong response: Panic, apologize, try to explain it away "Oh, that's weird, it usually works better than this..." [Kills credibility. Team loses trust.]

Right response: Turn failure into the lesson "Perfect! Stop right there. This is exactly what I wanted to show you. Look how confident it sounds while recommending a SQL injection vulnerability. This is why code review AI is a starting point, not a replacement for security expertise. Sarah, you spotted that immediately—AI didn't. That's the skill gap AI can't bridge."

Follow-up: Show the feedback loop "Now watch what happens when I tell it Sarah's concern." [AI corrects course when given expert input] "See? It's like having a junior developer who's read every programming book but has never been hacked. Useful for catching obvious issues, but you still need senior judgment for the subtle stuff."

Result: Failure becomes proof point instead of credibility killer

Quality Gates

Post-Introduction Validation Checklist:

• Aha-moment observed: Learner showed visible surprise/recognition during demo
• Emotional safety confirmed: Learner expressed concerns and received honest responses
• Hands-on completed: Learner personally used AI on their own task/content
• Rung mastery verified: Can explain current capability level to someone else
• Next action identified: Specific task they'll try with AI this week
• Limitation awareness: Can name at least one thing AI is bad at
• Recovery confidence: Knows what to do when AI gives poor output
• Ready for self-guided use: Asking "how can I..." instead of "can you show me..."
• Context awareness: Understands why this AI capability matters for their role
• Skepticism addressed: Core concerns acknowledged and honestly discussed

NOT-FOR Boundaries

Use AI Introduction Educator for:

• First-time AI experiences and onboarding
• Workshop design for mixed-technical audiences
• Overcoming emotional barriers to AI adoption
• Explaining agentic workflows to business stakeholders

Do NOT use for:

• Technical implementation: For building AI systems, use ai-engineer skill
• Advanced prompt optimization: For prompt engineering mastery, use prompt-engineer skill
• ML model training: For data science and model development, use ai-engineer skill
• Production deployment: For AI infrastructure and scaling, use agentic-infrastructure-2026 skill
• Developer tool training: For technical AI tool adoption by engineers, delegate to tool-specific documentation

Handoff signals:

• When they ask "How do I build..." → ai-engineer
• When they ask "How do I optimize prompts..." → prompt-engineer
• When they ask "How do I deploy..." → agentic-infrastructure-2026
• When they ask "How do I create training data..." → ai-engineer