Which AI app builder should a beginner start with: Replit, Lovable, v0 or Claude Code?

I built my first newsletter software in Replit and stayed there for seven or eight months. Today I use Claude Code and Codex most of the time. I don't think I could have skipped the first step. It would have been too confusing and too hard.

"Claude Code is like level three or four or five. You probably want to start easier."

Where should a non-coder start?

A hosted builder: Replit, Lovable, Bolt or v0. Think of Replit and Lovable as wrappers on top of Claude Code that make it easier. You describe what you want in plain English, it builds, you look at it, and if it didn't go the way you wanted, you tell it that in plain English too. Say "build me a to-do list software that does this, this and this" and it'll work. It probably won't work exactly how you want.

You're not an expert yet. This is the best time ever to start playing with these things. Back in early 2025, my technical partner's advice to me was to play with these tools for an hour a day for 30 days. My own estimate then was about 30 hours over three months to get comfortable. It's not a three-hour job, but it's not rocket science either.

Which builder makes the best designs?

My trick is to throw the same prompt into all of them and compare. I've run Bolt, Lovable, v0 and Replit side by side, and v0 was the winner about 90% of the time for me. Bolt once made a better-looking site, and in another round I rated it the worst, with errors. I designed my directory site's front end in v0 and it looked great, and I redesigned my newsletter software's dashboard in two prompts by dropping the code in. Their $20 plan is worth paying for.

None of them are magic on design. One day in January 2026 I spent hours with v0, Bolt, Lovable and Google AI Studio and none of them could design a page I liked. I was running out of credits on two of them, too.

How do I prompt a builder?

1. Have a chat AI write the first prompt. Tell ChatGPT, Claude or Gemini what you want built, ask it to ask you a few clarifying questions, then have it write a detailed prompt for the builder. When you give a little bit of info, that's why you get crap out of AI.
2. Build a container first, then features one at a time. When I was building in v0, the base layer, the general design and infrastructure, was its own build. Then each feature got its own build using that base prompt. Ask for multiple features in one go and it starts to screw up.
3. Expect a boring first version. The first thing you get is going to be boring and ugly. I tell it "this is boring, make it more visually appealing," and version two is much better.
4. Make it grade itself. After the first version, I ask:

Is this really the best you can do?
Please give yourself a review and tell me what you missed.

One caution on that first step. For pure website design, a super-detailed prompt can backfire because the builder tries too hard. On one landing page, I got further by giving the builder only the copy and letting it handle the look. More in how to stop AI slop.

What are the downsides of hosted builders?

• Cost. Replit seemed to run out of tokens really fast, so you end up paying more. One way to cut it: a bundle I've mentioned gives you a year of about 15 AI apps, including Replit, Bolt, Lovable and n8n, for about $200.
• Undo doesn't really undo. One of my biggest frustrations: I got far, loved it, then it screwed one thing up and I couldn't go backwards. The buttons say you can. It kind of really doesn't.
• Feature creep. We added too many features to the newsletter software on Replit, then had to remove a bunch because they confused non-technical users.
• A pretty shell isn't a product. My v0 directory looked great and had no businesses in it. The design was the easy part. The data was the hard part. When I designed apps in v0, a development team still had to do all the back-end work I had no idea how to do.
• Dead code piles up. My newsletter builder started about 16 months earlier in the early Replit days. When I asked a current model whether to upgrade or restart, it said start over. AI-built code collects cruft, and rebuilding on a current model can be cheaper than patching.

How far can a beginner actually get?

Further than you'd think, and not as far as the hype. You could vibe code a simple app in about 20 minutes. Getting to the 95% done mark is shockingly easy. Getting from 95% to 100%, where every button and flow actually works, is still super hard. I'm still struggling with it.

That gap is a business, by the way. In 2025 I said the biggest opportunity was an agency that finishes Bolt or Replit apps for people: "that thing you built, I can make it work for five grand or 10 grand." So many people build something that looks great and doesn't work.

When should I move to Claude Code or Codex?

When the thing you built turns into a real product for a bigger business. If it's simple, you can probably pull it off completely in Replit. Once it gets complicated, get off it. That's when I moved my newsletter software: we were trying to make it almost enterprise level, and Replit isn't built for that. Lovable is mostly prototyping. Claude Code is how you go deeper and make a real app you can go sell.

The step up is mostly getting used to working in folders on your computer, which takes about 15 minutes to understand. I cover the rest in how a non-technical founder builds software with AI.

What should I build first?

Something you'd use this week. Then ask the question I push every builder on my podcast to ask: would anyone else pay for this? A lot of people build things for themselves that are kind of helpful. The ones worth productizing solve a problem people would actually pay to fix.