You've tried ChatGPT, generated a few images, maybe even got a bit of code running. But as long as those stay isolated experiments, you have nothing to show. Things really click when you take a project from start to finish.

First projects rarely die from a lack of skills. They die because they're too vague, too big, or never finished. Here's a simple method to choose, scope and finish.

The four conditions of a good first project

1. It's useful, to you or to someone real

For yourself, a friend or family member, or a nonprofit: someone is waiting for the result. That gives you a reason to finish and honest feedback.

2. It draws on several skills

Text, visuals, a page (a small website), maybe even a bit of research or automation. Aim for at least three of these: that's where you learn to combine tools.

3. It fits in 6 to 10 hours

Split into sessions of an hour or two. Enough to produce something solid, not enough to get lost.

4. The result is visible

A page online, a document, a small tool. "I learned a lot" isn't something you can show. A link is.

Four ideas to get started

  • A website for a local nonprofit: introduction, opening hours, visuals, contact.
  • A travel guide: itinerary, places you check yourself, an illustrated cover.
  • A personal tutor: an assistant set up to help you review a subject, with exercises.
  • A fictional shop: an invented brand, product pages, a storefront page. No real sales.

If the tutor idea appeals to you, start by seeing how to study with ChatGPT without cheating: flashcards, quizzes and explaining things back.

None of these speaks to you? Follow our method to brainstorm ideas with ChatGPT: set your constraints, get a long list, then sort it with your own criteria.

The project sheet

Before opening any tool, write a one-page sheet with ten sections:

  1. Name of the project.
  2. Who it's for: who will use the result.
  3. Problem: what's wrong today, in one sentence.
  4. Visible result: what you'll show at the end.
  5. Deliverables: three to five concrete items.
  6. Tools: which ones, for what.
  7. Out of scope: what you won't do.
  8. Success criteria: checkable with a yes or a no.
  9. Sessions: how many, and what you do in each one.
  10. Risks: what could block you, and your plan B.

Not sure what to put under "Tools": an assistant, an image generator, an AI-assisted coding tool or an automation? The free mini-game Which AI for the job? trains you to sort everyday tasks into the right family of tools, in a few minutes.

A filled-in example

Name: The book club page. Who it's for: members of the neighborhood book club. Problem: the info (dates, book of the month, venue) goes around in messages and gets lost. Visible result: a web page online whose link can be shared. Deliverables: a home page with the book of the month; an introduction text; three visuals; a meeting calendar. Tools: an assistant, an image generator, an AI-assisted coding tool. Out of scope: no members' area, no forum, no online sign-up. Success criteria: Is the page online? Is the book of the month visible without scrolling? Did a member find the date of the next meeting on their own? Sessions: five sessions of an hour and a half. Risks: publishing it online gets stuck. Plan B: share the page as a document.

Every criterion can be checked with a yes or a no. "A nice-looking page" is not a criterion.

The "out of scope" line and the project that keeps growing

This is the most important line. Along the way, ideas show up: "what if I added a members' area?" Each one seems small. Together, they turn eight hours into a project that never ends.

When an idea shows up, you have two options: write it down in a "version 2" list, or remove something else to make room for it. Never add without removing. AI makes the trap more tempting, because producing feels fast. Checking, integrating and fixing, much less so.

Have a demanding project manager review your sheet

An assistant playing a strict project manager quickly spots anything vague. Copy this prompt:

You are a demanding project manager who coaches beginners. I'm giving you the sheet for my first AI project. I have 6 to 10 hours in total. Review it and tell me: 1) which success criteria can't be checked with a yes or a no, and how to rephrase them; 2) which deliverables are too big for the time available; 3) what should move to out of scope; 4) the risks I've forgotten. Be direct, don't congratulate me, and don't rewrite the whole sheet: just list the problems and your suggestions. <sheet> [paste your project sheet here] </sheet>

Keep your own judgment: accept the comments that make the sheet clearer or smaller, ignore the ones that make it bigger. More templates like this one are waiting in our ChatGPT prompt examples.

A session plan that holds up

For about eight hours:

  • Session 1: scope. Sheet, review, final version.
  • Session 2: the text. Writing and proofreading the content.
  • Session 3: the visuals. Generating and sorting (see generate an image with AI).
  • Session 4: the page. Putting it together with an AI-assisted coding tool (see coding with AI without being a developer).
  • Session 5: check and deliver. Criteria one by one, fixes, going live.

End each session with two lines: where you are, and what you'll do next.

Finish and show it

In the last session, go back to your criteria and answer honestly with yes or no. All yes? It's done, even if you can see improvements: those go into "version 2".

Then send the link to the person it's for with a single question: "Did you find what you were looking for?" Write down what you learned: it's the foundation of your second project.

Going further

This method is at the heart of Ovrelia's Workshop. The "Choose and scope" chapter has you pick a project, fill in your sheet and get it reviewed by an assistant playing a demanding project manager. It's free and runs in your browser, no account needed: scope your project now. And to check where you stand, try the friend challenge: ten timed questions about AI, and a score for your friends to beat.

Want a very concrete first project? Meal planning with AI takes about fifteen minutes and gives you something you'll use the very next week.

Frequently asked questions

What AI project should a beginner choose?

Choose a project that's useful to you, someone close to you or a nonprofit, that draws on at least three skills (text, visuals, a page, automation), fits in 6 to 10 hours and has a result you can show. A nonprofit's website, a travel guide, a personal tutor or a fictional shop are good starting points.

Do you need to know how to code to do a project with AI?

No. Assistants and AI-assisted coding tools let you produce a simple page without being a developer. What matters most is describing precisely what you want and checking the result.

What is the "out of scope" line for?

It lists what you won't do in this project. It stops you from adding features as you go and turning a few-hour project into one that never gets finished.

How do I know if my project is a success?

From the start, write criteria you can check with a yes or a no, for example "the page is online" or "a user finds the information on their own". At the end, go through them one by one: if they're all yes, the project is done.