13 · Check your understanding

1. What's the right order when turning a validated need into an AI product?

Pick a cool AI tool first, then find a use for itNeed first, AI second — aim AI at the job it can do reliably, with you accountableBuild the biggest version possible to look impressiveSkip the customer and trust the model

2. Your idea is 'when a request comes in, summarize it and draft a reply.' Which build path fits the first version best?

Build a custom model from scratchNo-code / low-code with an AI step in the middleHire an engineering team immediatelyNothing — it can't be built

3. Why does 'just a text box on a model' usually fail as an AI product?

Models are too cheap to runBecause the product is what you wrap around the model — expectations, control, sources, and a graceful failureBecause users prefer no instructionsBecause AI is always right

4. What does 'human-in-the-loop' mean, and when is it most required?

A human writes everything; AI is never usedA person checks/approves AI before it ships — most required for anything a customer relies on or that's irreversibleThe AI supervises the humanIt only applies to big companies

5. Why can an AI feature that's 'free' in testing become a real bill at scale?

APIs charge a one-time feeYou pay per token (input + output) on every call, for every customer — so cost grows with usageTesting always costs more than productionAI tools never charge for usage

6. In pricing an AI product, what sets the floor and what sets the ceiling?

The floor is whatever competitors charge; the ceiling is unlimitedYour cost-to-serve sets the floor; the value you deliver to the customer sets the ceilingThere is no floor; price as low as possibleThe model's name sets both

7. Which approach to a possible AI wrong answer BUILDS trust?

Always answer confidently, even when unsureMake uncertainty visible, verify what matters before it ships, and let customers correct/report itHide that AI was involvedRemove all ways to reach a human

8. What does the FTC say about claims that your product 'uses AI' or about what your AI can do?

You can say anything as long as it sounds impressiveAI performance claims must be truthful and substantiated; 'the AI said so' is not a defenseOnly claims about price are regulatedClaims don't matter for small businesses

9. Can you copyright a logo you got from a single AI prompt, according to the U.S. Copyright Office?

Yes — anything AI makes for you is automatically yoursPurely AI-generated output isn't protected; copyright needs human authorship, and prompting alone isn't enoughYes, if the prompt was detailedOnly if you paid for the AI tool

10. What's the goal of a small first launch of an AI product?

To go viral immediatelyTo learn fast and cheap with real users whether it solves the problem — then scale what's provenTo avoid ever talking to customersTo ship the most polished version possible before anyone uses it