Course A · Module A1 Foundations · Lesson 1 of 17

What AI Can Actually Do For Your Business

Before you automate or build anything, you need an honest picture of what this tool is — and isn't.

~12 min · No tools needed yet · Bring: one task you did three times last week

You've heard AI will either run your business or ruin it. Both are wrong. The founders who win with AI aren't the ones who believe the hype — they're the ones with an accurate mental model, who know exactly where to point it. That's this lesson's single win: by the end, you'll be able to look at any task in your business and say "AI belongs here" or "it doesn't" — and be right.

The mental model: a brilliant, forgetful intern

Picture the smartest intern you've ever met. They've read almost everything, write fast, never get bored, and work for pennies. But: they have no memory of yesterday, they've never seen your business, they can't check anything live, and when they don't know something they'll confidently make it up rather than admit it.

That intern is an AI model. Every strength and every danger below falls out of this one picture.

✓ Great at

  • Drafting & rewriting (emails, posts, listings)
  • Summarizing long things into short ones
  • Sorting & tagging (which bucket does this go in?)
  • Turning messy notes into clean structure
  • First drafts of anything repetitive

✗ Bad at

  • Facts, numbers & dates (it hallucinates)
  • Anything needing live data it can't see
  • Remembering past chats on its own
  • Final say on high-stakes / legal calls
  • Knowing your business unless you tell it

Notice the pattern: AI is strong on language and judgment-light transformation, weak on truth and memory. You get the strengths for free; you design around the weaknesses. That's the whole game.

The Automatable Task Test

Here's the practical tool. Take any recurring task and run it through four questions. The more "yes" answers, the better an automation candidate it is.

Does this task…
  1. Repeat? You do it weekly or more. (Rare tasks aren't worth automating.)
  2. Follow a rough rule? You could half-explain it to a new hire.
  3. Live mostly in text? Reading, writing, sorting, summarizing.
  4. Survive a wrong draft? A mistake is annoying, not catastrophic — or a human checks it before it ships.

Four yeses — automate it soon. Two or three — automate it with a human checkpoint. Fewer — leave it to a person for now. You'll use this test in Lesson 3 to pick the one workflow you'll automate over this course.

Check yourself

Answer before opening. Recalling it now is what makes it stick.

A client asks you to draft a friendly reminder email for overdue invoices. Good fit for AI?

Yes. It is text, it repeats, it follows a rough rule, and a wrong draft is harmless because you read it before sending. Textbook candidate.

You ask AI for the exact total of last month's revenue from memory. Good fit?

No. That is a live fact and a number — exactly where the "forgetful intern" confidently makes things up. Pull the figure from your records, not from the model.

Why does giving the model your brand-voice examples in the prompt improve results?

Because it has never seen your business. It only knows what you put in front of it. Context you supply is the difference between generic and on-brand.

Read this next (primary source)

Generative AI: Prompt Engineering Basics — IBM (Coursera). The opening module reinforces this same "natural-language-as-instruction" model, built for non-technical readers. ~20 min. See RESOURCES.md for more.

I'm your teacher — ask me anything. Stuck on whether one of your tasks passes the test? Not sure what "hallucination" looks like in practice? Reply and I'll walk through your specific example. That back-and-forth is where the real learning happens.

New words this lesson are defined in the course glossary — AI model, prompt, context, hallucination. Keep it open as you go.