The process of going from idea to a working prototype you can ship has never been this fast, and it's all because of AI. While you once needed to learn the fundamentals of programming yourself by taking a few courses, or hiring someone who has already done the learning, all it takes now is a prompt to begin building. Sure, you'll likely keep finding vibe-coded apps that leak user data (without even looking for them).

But is that a reason to not explore what these tools can actually do when used thoughtfully? Vibe-coding random tools and apps that help my workflow has quickly turned into the most fun I've had with AI, and I tell people around me left and right to give them a serious try. Four of the biggest names you'll find in this space are Claude Code, Codex, Lovable, and Replit. I've been using all of these tools extensively, and I wanted to see how they stack up when you put them to work on the same project. So, of course, that's exactly what I did.

I used all four tools to build the same collaborative app

One prompt, four very different outcomes

Claude Code and Codex are both primarily terminal-based tools, designed to work within folders local to your machine. They're largely developer-focused, but once you get over the slight learning curve, I'd argue they're accessible to anyone. Lovable and Replit, though, have no real learning curve. You just open the tool's website in your browser, create an account, and start prompting. Unlike Claude Code and Codex where you get the result at the end, Lovable and Replit let you watch your app take shape in real time, which makes it extremely easy to course-correct along the way. I've been using all four tools in different capacities, and I've even pitted some of them against each other before. But for this experiment, I decided I'd ask all four to build something I've never really asked them to before: a real-time collaborative app.

This way, I'd be able to test how they handle something that actually requires multiple systems working together. The UI comes into play too, of course, but I wanted more so to analyse how each tool handles what's happening under the hood. Something I've mentioned in a lot of the articles I've written comparing these vibe-coding tools is that I prefer using a somewhat vague prompt to see how much each tool can figure out on its own. AI tools have gotten better at following instructions, so if I spell out every detail, all I'm really learning is whether they can follow directions. That said, I did make sure my prompt had every detail needed to understand the vision. Here's the exact prompt I used across all four tools:

Claude Code and Codex built the app well

They couldn't ship it though

Now keep in mind that I'm testing these tools from the perspective of a non-developer who just wants to go from idea to something I can share and use. Claude Code was the tool that finished building the quickest, and Codex followed behind it closely. That said, there's a reason they managed to finish earlier than Replit and Lovable, which have always been faster in my previous testing.

Claude Code and Codex only had to build the frontend, which is the part you actually see. They skipped the backend entirely. They didn't set up a database, a hosted server, add authentication — nothing. Both tools used something called localStorage and BrodcastChannel to simulate real-time collaboration, essentially meaning that the app only worked across tabs on the same browser. You couldn't really open it on your phone or send the link to a friend and see how it held up in real-time.

Upon asking each tool how I could deploy it and get it moving, both handled me a list of steps to do myself. Codex pointed me to two separate tools (Supabase for the database and then Vercel to deploy it), while Claude Code went a step ahead and created a GitHub repo. However, it did tell me to go deploy it on Render myself.

Codex's version is a shared weekly meal-planning board. At the top, you enter your name and either join an existing board or create a new one. Each board has its own shareable URL, so anyone opening the same link lands in the same planning space. On the left is the meal library, where you add meals with a name, optional cook, dietary restriction tags, and ingredients. Those meals become draggable cards.

In the center is the week board where you drag meals from the library into any day, and once a meal is scheduled, you can reassign who's cooking, move it to another day, or unschedule it. On the right is the grocery list, which updates automatically based on everything on the board. There's also a notes area and a dietary summary that gives a quick count of restriction-tagged meals across the week. For collaboration, people on the same board can see who's online, and cursor indicators show where others are active.

Claude Code's version follows a similar structure but takes a slightly different approach. You start on a join screen where you enter your name, and once you're in, you see three panels. The left side has the meal library and a form to add meals with ingredients, quantities, and dietary flags. The center is a weekly board with three slots per day (breakfast, lunch, and dinner), which gives the planning a bit more structure than Codex's open-ended day columns.

The right panel is the grocery list, which auto-populates based on what's on the board and includes checkboxes for ticking off items as you shop. To assign a cook, you click on any placed meal and a modal pops up showing everyone currently online. The presence bar at the top shows colored badges for connected users, and other people's cursors appear on your screen as colored arrows with their name attached. Everything syncs live!

In both cases, the final versions were impressive. Codex had the better UI, though it felt a bit...too much. Claude Code's UI felt a bit old-school, but wasn't bad by any means. While both the tools worked great and didn't really need any follow-up prompting to get to a functional state, neither one really gave me something I could begin using right away (which

was the goal of this entire experiment).

Replit and Lovable handled everything and were the easiest to use

Including the parts I didn't ask about

Replit and Lovable handle the entire stack for you, including the frontend, backend, database, and deployment without you really lifting a finger. I sent the exact same prompt as above to both, and they first built a working prototype within minutes. I could preview it within the tool's interface, and play around with it in real time. When I was ready to see its collaborative features in action, I just hit publish! No Supabase, no Vercel, no Render, no GitHub repo.

Replit's version, which it named "Kitchen Sync" on its own, took about eight minutes to build and came with a production database already connected. You set up a profile with a name and color, create a board for your household, and share the link. Anyone who opens it joins the same board. From there, it's drag-and-drop meal planning. You assign meals to days, pick who's cooking from the list of people actually on the board, flag dietary restrictions, and watch the grocery list build itself.

You can see who's online and watch cursors move in real-time. Personally, I liked Replit's UI the most too. It was minimal, aesthetically appealing to the eye, and looked like a tool I'd actually want to use.

Lovable's version took a similar approach but leaned harder into the collaborative side. The meal library, weekly board, cook assignments, and grocery list all sync live across everyone on the same board. It even lets you manually add items to the grocery list and check things off, and those changes sync too. The presence bar shows colored avatars and a live count of who's online, and when you assign a cook, it suggests names of people currently on the board. Lovable was also upfront about what it hadn't built. It told me live cursors weren't included yet but offered to add them.

Sharing it with others was fairly simple too. There's a Share button in the header that copies the board link to your clipboard, and once published, anyone with the URL can jump in.

Replit wins this round

I've tested Replit before in these kind of head-to-head comparisons, and this is the first time it's impressed me so much. In previous tests, it always felt a step behind Lovable. This time, it pulled ahead. That said, Lovable's output wasn't behind by any means either. Replit just had the edge in UI polish!

Claude Code and Codex both produced impressive apps. I want to be clear about that. The reason they fell behind in my testing was simply beacuse they expect you to figure out how to ship it yourself. And for someone who just wants to go from prompt to shareable link, that gap is everything.