Every pitch deck has AI in it now. Every startup claims to be “AI-powered.” Most of it is nonsense.

I’ve been building startups for 15 years. Used AI seriously for two. Here’s the guide I wish someone had given me.

What AI Actually Does Well Right Now

Text generation: Writing drafts, summarizing documents, answering questions from a knowledge base. This works and is immediately useful.

Code assistance: Not writing whole applications, but helping with specific functions, debugging, and boilerplate. Real productivity gains here.

Image creation: Marketing graphics, placeholder images, quick visualizations. Not perfect but fast.

Classification and sorting: Categorizing support tickets, tagging content, routing requests. Surprisingly reliable.

Transcription: Speech to text is basically solved. Fast and accurate.

That’s mostly it. Everything else is either unreliable, expensive, or both.

What AI Does Poorly Right Now

Decisions involving real money: Don’t let AI approve expenses, set prices, or choose vendors. It hallucinates and has no skin in the game.

Anything requiring current information: Training data is old. Without web access, AI confidently discusses things that no longer exist.

Nuanced customer communication: It can draft; it shouldn’t send. Tone-deaf AI responses damage relationships.

Complex reasoning chains: Multi-step logic where each step depends on the last. AI loses the thread.

Anything legal or compliance-related: Please don’t. Just don’t.

Where to Start

If you’re just getting into AI for your startup:

Week 1: Get Claude Pro from Anthropic or ChatGPT Plus from OpenAI. Use it for everything. Writing, research, brainstorming. Get a feel for what it can do.

Week 2: Identify three repetitive tasks in your business. Try using AI to speed them up. Time the before and after.

Week 3: Pick the one task where AI helped most. Build a simple workflow around it. Document the prompts that work.

Week 4: Evaluate. Did it save time? Was the quality good enough? Is it worth continuing?

Don’t build custom AI products. Don’t hire AI agencies. Just use the tools for a month and see what happens.

The 80/20 of AI for Startups

80% of the value comes from:

• Drafting content faster
• Summarizing long documents
• Answering internal questions
• Basic customer support

These require no technical skills. No custom development. Just $20/month and some practice with prompts.

The other 20%—custom models, complex integrations, AI products—needs significant investment and expertise. Leave that for later.

Red Flags to Watch For

“We’re building AI into everything”: Unfocused. Probably burning money.

“Our AI learns and improves automatically”: Usually means nothing. Or means unpredictable behavior.

“We use proprietary AI technology”: Almost certainly just using GPT/Claude with custom prompts. Which is fine, but not special.

“AI replaces X entirely”: Nope. Augments, maybe. Replaces, no.

What I Tell Founders

Use AI where it’s obviously useful. That’s content, summarization, and simple automation.

Don’t let FOMO drive decisions. Most AI projects fail. The ones that succeed solve specific, boring problems.

If you’re not sure whether you need AI, you don’t need AI. When you need it, you’ll know.

And if someone’s pitching you an AI solution that sounds too good to be true, it is. Every time.

The founders who win aren’t the ones with the most AI. They’re the ones who use AI thoughtfully on the things that matter.

That’s the whole guide. Sorry it wasn’t more exciting.