Working with AI
Alice asks her AI helper, Bit, for a bakery website. Bit builds a spaceship.
After this lesson, you'll be able to give your AI a clear task and decide what evidence you need before accepting its work.
How you work with the AI decides what you pay. A vague prompt burns tokens and hours on rewrites. A change too big to review lets bugs slip through to users. The wrong model either spends money on an easy task or guesses on a hard one. Clear, small, checked tasks on the right model make the same AI cheaper and faster, and its mistakes easier to catch.
Ask the AI so it builds the right thing: smallest version first, a clear shape for the answer, a feature thought through before you prompt. Keep each change small, know what the AI can see and what it costs, and check what it wrote. The AI is fast and sounds sure even when it's wrong, so a change that looks fine can still be broken. These habits catch that before your users do.
What this lesson covers
- 1What to build first: MVP & scope
An MVP, a minimum viable product, is the smallest version you can show real people to test if your idea is worth building.
- 2Be specific: say the shape
Say what's wrong, where in the app, and what you want back.
- 3Design a feature before you prompt
Before asking an AI to build something, think through five steps in order: requirements, flow, data, requests and screens.
- 4Edge cases & the failure path
The happy path is when everything goes right.
- 5Small tasks & a definition of done
Break a feature into small changes.
- 6Context, tokens & cost: what the AI can see
Context is what the AI can see right now: your request, selected files and part of the conversation.
- 7Review what the AI wrote
Don't take the AI's word that something works.