Day 1 - Discover Your AI Idea

Stop Overthinking It

LLMs predict the next word. That's it. That's the magic.

You feed it a bunch of text. It figures out which word comes next. Then it predicts the next word after that. And again. And again. The result feels like intelligence, but it's really just very good pattern matching trained on most of the internet.

This is enough to be incredibly useful. You don't need to understand transformers or attention mechanisms. You need to understand what these tools can do for real people with real problems.

The 3 Types of AI Tools

Chatbots - You talk, they respond. ChatGPT, Claude, Gemini. Great for writing, research, brainstorming, code, analysis. This is where most people start.

Generators - You describe, they create. Midjourney and DALL-E for images. Suno for music. Runway for video. These turn words into media.

Agents - You instruct, they act. These tools don't just answer questions - they take actions. They book meetings, write and send emails, update spreadsheets, deploy code. This is where AI gets truly powerful.

Hands-On in 5 Minutes

Open Claude (claude.ai) or ChatGPT. Try these two tasks right now.

Task 1 - Write a LinkedIn post:

Paste this into Claude:

Write a LinkedIn post about why I'm leaving my corporate job to build an AI startup.
Tone: honest and reflective, not cringe. No hashtags. No "I'm thrilled to announce."
Keep it under 200 words. End with a question to drive comments.

Task 2 - Write a cold email:

Write a cold email to the owner of a local plumbing company.
I'm selling an AI tool that answers their missed calls and books appointments automatically.
Keep it under 100 words. Lead with the problem (missed calls = lost revenue).
No jargon. Sound like a real person, not a marketer.

Notice the difference between asking "write me a LinkedIn post" and giving it specific constraints. That's prompt engineering.

Prompt Engineering Basics

Four rules that cover 90% of what you need:

1. Be specific. "Write a blog post" is bad. "Write a 500-word blog post about tenant screening for small landlords" is good.
2. Give context. Tell it who you are, who the audience is, what the goal is.
3. Show examples. "Write it in the style of this example: [paste example]" works better than describing a tone.
4. Assign a role. "You are a senior copywriter at a B2B SaaS company" changes the output dramatically.

Boring Problems Make Great Startups

The best AI startups don't solve flashy problems. They solve boring ones really well.

• Invoice processing for construction companies
• Insurance claims summarization
• Tenant maintenance request routing
• Dental appointment reminders
• Freight quote comparison

Nobody posts about these on Twitter. They just quietly make money.

The formula: Boring niche + AI = defensible business.

Why? Because the big companies (OpenAI, Google, Microsoft) will never build "AI invoice processing for HVAC contractors." It's too small for them. It's perfect for you.

How to Spot AI Opportunities

Go where the complaints are:

• Reddit and forums. Search "[industry] + frustrated" or "[industry] + waste of time." People tell you their problems for free.
• Find the hated spreadsheet. Every company has one spreadsheet that someone spends 4 hours a week updating manually. That's your product.
• Look for the copy-paste job. If someone copies data from one system to another, AI can do it.
• Listen for "I wish someone would just..." That sentence is a product brief.
• Watch for the 3-day turnaround. If something takes 3 days that should take 3 minutes, there's a gap.

7 Categories of AI Applications

When brainstorming, think across these buckets:

1. Content and Communication - Blog posts, email sequences, social media, internal memos
2. Customer Experience - Chatbots, personalized recommendations, automated support tickets
3. Data and Analytics - Report generation, trend detection, anomaly alerts
4. Operations and Workflows - Document processing, scheduling, inventory management
5. Knowledge Management - Internal search, FAQ bots, onboarding assistants
6. Creative and Design - UI mockups, product photography, video editing
7. Decision Support - Risk scoring, pricing optimization, hiring screening

Pick the category where you have domain knowledge. That's your unfair advantage.

How to Talk to Users Without Lying to Yourself

This framework comes from Rob Fitzpatrick's book The Mom Test. The core idea: your mom will lie to you. She'll say your app idea is great because she loves you. Most people do the same thing when you ask "would you use this?"

The rules:

1. Talk about their life, not your idea. Don't say "I'm building an app that does X, would you use it?" Say "Tell me about the last time you dealt with [problem]."

2. Ask about the past, not the future. "Would you pay for this?" is worthless. "How much did you spend on this problem last year?" is gold.

3. Ask about specifics, not generalities. "How do you usually handle invoicing?" is better than "Is invoicing hard?"

Good questions:

• "What do you currently do when [problem] happens?"
• "How much time do you spend on this per week?"
• "What have you tried to fix this? Why didn't it work?"
• "Who else in your company touches this process?"
• "If you could wave a magic wand, what would change?"

Bad questions:

• "Would you use an app that does X?"
• "Don't you think AI could help with this?"
• "How much would you pay for this?"

The difference: good questions give you facts. Bad questions give you opinions. Build on facts.

Your Day 1 Assignment

Step 1 - Brainstorm

Pick 3 problems from the categories above. For each one, write a 2-sentence pitch.

Example: "Dentists lose $50K/year from no-shows. An AI tool that sends personalized reminder sequences and reschedules appointments automatically would cut no-show rates in half."

Step 2 - Score

Rate each idea:

Step 3 - Validate

Take your #1 idea and talk to 3 people who have the problem. Use Mom Test questions. Write down what they actually said - not what you hoped they'd say.