What is the best first project?
The best first project is small, visual and about something you already know well, so you can check every word yourself. A one-page site, not an app: you stay the authority, and the model builds without becoming it.
Your first coding project does not have to be your first experience of AI. I have worked with generative AI models since they became commercially available. During my doctorate I used them to help summarise and navigate a large body of interview transcripts, and to move back to the passages that mattered more quickly. That taught me something I have not unlearned since: a convincing summary is not necessarily an accurate one. I still had to return to the transcript, check the original context, and stay responsible for the interpretation.
So when I first built software of my own, I was not starting from nothing. I was building on years of learning where these tools help, where they mislead, and what has to stay a matter of human judgment. Vibe coding with veracity rests on exactly those habits, and it left me with a question worth answering for anyone starting today: what makes a good first project?
Why is now a good time to begin?
Generative AI is advancing extraordinarily quickly. Things that were difficult a short time ago are becoming possible through an ordinary conversation with a model. That pace can make it tempting to wait until the technology settles down, and it is unlikely to do so soon.
This is a good time to dip a toe in, not because you need to master every model, but because you can start building habits that will outlast any of them:
- Describing what you want clearly
- Giving the model sensible boundaries
- Noticing when something does not look right
- Checking claims against their sources
- Testing what has been built
- Staying accountable for what you publish
The tools will keep advancing. Starting small now lets your judgment advance alongside them.
So what should the first project be?
Build a one-page website about something you already know well. For example: five places you would take a visiting friend, your five favourite local walks, a beginner’s guide to keeping houseplants alive, or what you wish you had known before medical school. My version would be My Watford: five places I would take a visiting friend, each with a photograph, a short personal note on why I chose it, and a link to its official page.
This is a better first project than an app, a chatbot, a login system or a database, and for one reason: you can verify it yourself. You already understand the subject. You know what the page should say, you can see when it looks wrong, and you can test every link and button. That is the whole game. The best first project is the one where you, not the model, are the authority.
How small should it be?
Small enough that nothing on it is beyond your own checking. A good first project is:
- Small enough to finish in a session or two
- Personally useful, or worth showing one real person
- Visual, so mistakes are easy to see
- Built on information you understand
- Free of accounts, payments and private data
- Free of factual claims you cannot check
- Easy to publish as a simple website
You are not building the final version of a grand idea. You are making a safe place to learn how you and the model work together.
But didn’t you start with a dashboard?
I did. My own first build was not a one-page site at all: it was a personal dashboard, planned in an evening and built the next morning, work that once took months done in a weekend. So why not send you the same way?
Because it was not my first time with these models. The doctorate had already taught me where they help and where they mislead, so I could take on something with live, moving data and still tell when it was wrong. That judgment is the thing you are here to build, and you build it fastest on ground you already know like the back of your hand. The dashboard can come later. Starting small is not starting timid; it is starting where you can tell right from wrong.
What would the brief look like?
Close to this:
Build a one-page website called “My Watford: five places I would take a visiting friend.” Give it a welcoming introduction and five place cards. Each card has a photograph, the name of the place, my short personal note and a link to its official page. Use only the information I provide; if something is missing, mark it clearly rather than inventing it. Make it warm, calm and easy to read on a phone. Do not add accounts, reviews, booking tools or a map yet. I will check every word, photograph and link before publishing.
That is already most of a good brief: a clear goal, a tight scope, a firm “not yet”, and a promise to verify. The next session turns that into a repeatable shape you can reach for every time.
What does a first project like this teach?
The whole loop, in miniature. You describe what you want, read the first version, correct the words and the look, test the links and the phone layout, publish, and then make one change after it is live. Every habit the later sessions lean on, you meet here first, on something too simple to hide a mistake in.
The deeper lesson is the one to carry out: your first project should use your knowledge, so the model can build the software without becoming the authority. That single sentence is the whole of veracity, practised on your first day.
How do I choose my own?
Write down three ideas and score each against these:
- Do I understand the subject myself?
- Can it fit on one page?
- Can I check every claim?
- Can I finish a first version in a session or two?
- Would I show it to one real person?
- Can it work without accounts, payments or private data?
- Would a mistake be easy to spot and fix?
The idea with the most yeses is probably where to begin.
So, what is the best first project?
Small, visual, and about something you already know well: a single page you could show one real person, with no accounts, no payments and no claim you cannot check. Not because you are not capable of more, but because the first project is where you learn how you and the model work together, and that lesson lands cleanest on ground you can judge. Choose the subject you know like the back of your hand. Let the model build the page. Keep the checking, and the final word, for yourself.