There's a question on V2EX that a lot of people are quietly asking: someone with a bit of frontend background, honest about being a weak self-learner, wanting to start with AI-assisted coding and asking for the right tutorial or video to begin with. They even begged the comments not to say "just ask the AI." It's a fair ask, and it deserves a real answer instead of a link dump.
Here's the thing that stood out to me reading the replies. The most useful one wasn't a course. It was a builder saying, roughly, that you have to actually do it first — that watching tutorials and reading books is worth less than shipping something with your own hands, and that a lot of "I don't know how to start" resolves itself the moment you start. Another commenter made the same point from the AI angle: asking the AI works, but only if you can tell whether it's doing sensible work or just flailing, and you build that judgment by shipping and watching what breaks.
That matches a pattern across the rest of the board. One person built a whole lesson-prep system over a month of AI chats — it works, it's in use, and the mess only became visible because it was live. A new grad put an AI video site online with signup, payment, and deployment all wired up, and called it a practice project to see if he could take something from zero to launched. The recurring lesson from the independent-dev threads is blunt: an MVP that real people can touch beats a half-year of adding features you never ship.
The trap: collecting tutorials instead of shipping one thing
If your self-teaching is weak, more tutorials make it worse, not better. Video after video gives you the feeling of progress without the friction that actually teaches you. The friction — the render bug, the endpoint that returns the wrong shape, the CORS error — is the curriculum. You only meet it by putting something online.
So the move for a frontend-literate beginner is to pick the smallest possible thing that runs end to end and ship exactly that. Not an app. Not a platform. One slice: a page in front, one function behind it, both hosted, both reachable from a browser. Once that loop is real, everything else is just adding more of the same.
The buildable loop: HTML page → Python endpoint → back to the page
You already know HTML, so start where you're strong. Build a single static page — a form, a button, a text box, whatever your idea needs — and get it hosted so it has a real URL. On VicroCode you can run HTML online and publish it, so step one is a page that actually exists on the internet rather than a file on your desktop. That alone is more than most tutorial-watchers ever finish.
Then give that page one thing to call. Write a small Python function that takes the input from your page, does something with it, and returns a result. You can run Python online and expose it through API Endpoint Hosting, which means your HTML can fetch from a real endpoint instead of you standing up a server yourself. The whole loop becomes: page collects input, calls the endpoint, endpoint returns data, page shows it. That's a complete round trip, and it's small enough to finish in one sitting.
Keep the Python trivial at first — reverse a string, sum some numbers, format a date. The point isn't the logic, it's proving the wire works. Once the round trip runs, you swap the trivial function for whatever you actually wanted, and you're building. If your idea needs to remember things between visits, VicroCode has SQLite databases you can add later; if it needs a model, the Model Center APIs cover models already on the platform. But don't reach for those on day one. Ship the loop first.
Where the AI actually helps here
This is where vibe coding earns its keep, and where the V2EX advice about "knowing whether the AI is right" becomes concrete. Use an AI coding assistant to write the fetch call in your page and the handler in Python — that's exactly the kind of boilerplate it's good at. Your job is to hold the boundary: does the endpoint return the shape the page expects, does the error case do something sane, is the AI inventing a library that doesn't exist. You learn that judgment fastest on a loop this small, because when something breaks there are only a few places it can be. If you want to go deeper on how to work with the assistant rather than just poke at it, the AI coding material is a better use of time than another generic intro video, because you'll be reading it with a running project in front of you.
The honest boundary
One straight answer so you don't waste an afternoon: VicroCode covers the HTML, Python, and hosted-API pieces of this loop, plus file management, SQLite, LanceDB knowledge bases, and publishing. It does not cover other stacks. If your plan involves a Flutter app, a Node backend, a particular cloud provider, or a framework outside Python, that's outside what I can point you to here, and you'd be fighting the tool instead of learning from it. For a first project, that limitation is actually a gift — it removes a hundred choices you're not ready to make and leaves you with the one loop that matters.
So, to the person who asked: stop looking for the perfect tutorial. Pick one tiny idea, build the HTML page, wire it to one Python endpoint, get both hosted, and click the button. When it works you'll have learned more than any video could teach, and when it breaks you'll have your first real question to ask — with actual context, which is the only kind of question worth asking an AI anyway.