I’ve been building a fairly ambitious product on v0. The first stages were incredible: fast UI generation, clean scaffolding, and a real sense that I could build the entire platform here. I wanted to stay fully inside the Vercel ecosystem, use the AI SDK, test models, and keep everything under one roof. I’ve invested time, credits, and genuine excitement.
But once the app grew into real complexity: multiple roles, cross-updating flows, state syncing, notifications, and interconnected dashboards, things started to break down. Data began behaving like spaghetti, actions didn’t propagate cleanly across user types, and the generated code became harder to maintain or reason about. v0 feels amazing for early momentum, but fragile when you try to turn that momentum into production-level structure.
I also acknowledge that maybe I’m reading v0 the wrong way. Maybe building a full multi-role application isn’t the intended use case at this stage. If that’s true, I’d genuinely like to know, so I can set the right expectations and choose the proper development path.
If anyone has gone through this moment, where v0 gets you far, but not quite far enough, I’d really appreciate your guidance. I want to stay inside Vercel, but I need clarity on whether v0 can realistically support a project of this scale.
And if this ever reaches Guille Rauch: lo escribo con respect y con esa mezcla de orgullo que compartimos los que venimos del mismo lugar. I want to keep building here, but I also need to understand the limits.
Your concerns are valid, but I’ll point out that the problems you outline (complex apps looking like spaghetti) are also very common in handwritten codebases, and they come from constantly adding new code without refactoring the existing code first
Once you have the basis for the product down and you see common patterns and use-cases, you can come up with a clean architecture for the code that will scale as you add new features. You can document code patterns so that new features are implemented the same way as old features
I’d like to see v0 (and all coding agents) get better at doing this automatically, but the state of the art coding agents still require some direction here.
If you’re using v0 as a “vibe coder”, where you don’t pay any attention to the code that’s being created, then you may be able to finish some apps before it gets too complex, but not all of them
Instead if you use the git workflow and create PRs and review each one, you can keep the code quality high and continue in v0 indefinitely
with git sync, you can do certain tasks (like codebase refactors) in an IDE and bring the changes back into v0 for further development
v0 is built in v0, but not the whole app. We have a semi-functional UI only clone that the design team uses as a template for new design mockups, which are then brought into the main codebase (in an IDE) after testing
It is our goal though to increase the capabilities of v0 and the v0 agent and expand what you can do entirely inside it. I’m not sure how long you’ve been using v0, but the things you can build now compared to what you could do 6 months ago is night and day and it’s only going to get better
And as you mentioned in the rest of your message, each AI coding agents out there is going through ups and downs and can only do so much (for now).
If one thinks about how far we have come with just AI in general in the short time….. its insane. So coding agents are also moving and catching up as well.
I think some that excelled prior have slowed, but surely they will pick up speed again. Soon it will be tools of preference, or what the company uses, or what fits your workflow and output needs.
And as you mentioned, where v0 was vs today… is crazy good. I mean no lie, when I first tried it ages ago, i was NOT a fan. I moved to Lovable right away. but about 2 months ago I wanted to give it a go again. So I started two projects simultaneously in Lovable and v0. Whatever prompt, edit, manual refinement I put into one, I would put into the other, so it was always a fair comparison. And well….. here we are. v0 just hit the marks and excelled in so many areas. I have canceled my subscription with Lovable and have had Premium with v0, and have been very happy sense.
And no shade to Lovable at all. They are a good company, great team (friend of mine is on it), and Swedish (same with me) so the love will always be there. But I go with what my gut says works the best “for me”.
Currently I would suggest to be explicit in the prompt you are giving, like the code structure you want it to follow and use the plan mode in v0 (believe me it cooks in that mode).
I second this. Plan mode alllllll the way. Rarely have any faults or fixes to make. Plus you can confirm what will happen before it does and it’s a fraction of the “price”.
I’ve hit this exact wall with v0 on a multi-role project. You’re not reading it wrong - v0 scaffolds brilliantly but starts to crack when you need state syncing across user types, real-time notifications, and cross-updating flows. Those features are just hard to maintain in generated code.
The pattern that works: let v0 handle your core business logic and UI scaffolding, but swap in stable components for the infrastructure-heavy stuff like notifications and user-to-user interactions. Web Components that handle their own state and sync independently won’t break when v0 regenerates code. Weavy’s one option (real-time notifications, activity feeds, role-based interactions in about 10 minutes), or you could build those pieces separately and keep them outside v0’s generation scope.
What specific flows are breaking down for you? The cross-role updates or the notification system?