Every couple of months someone sends me the same message: “is nobody using Lovable and v0 anymore? did they go out of fashion?”. It’s the wrong question, and I get why we ask it: when a tool stops showing up in your timeline you assume it died. But these didn’t die. Quite the opposite — they’ve never been so expensive, so well-funded, or so widely used as they are now.
The interesting question isn’t whether they went out of fashion. It’s a more uncomfortable one: now that the magic-demo hangover has worn off, are they actually worth it for what they promised? And there the answer isn’t the one marketing sells.
the numbers aren’t those of a fading fad
Let’s start by burying the “nobody uses them anymore” myth. The numbers say exactly the opposite.
In December 2025 Lovable closed a $330M Series B led by CapitalG and Menlo, at a $6.6 billion valuation — more than triple in roughly five months. It crossed $100M ARR faster than any software company in history, OpenAI and Cursor included, and reports more than 8 million registered users with around 100,000 new projects a day.
Bolt, from StackBlitz, went from zero to $40M ARR in about six months and boasts 5 million registered users. v0, from Vercel, is still the front door for half of product Twitter. The whole vibe coding market was estimated at $4.7 billion in 2026, with 63% of users being non-developers.
So no: they didn’t go out of fashion. If this were a bubble, it’d be a big one and still inflating. Which makes the next question more interesting, not less.
the billing model has an uncomfortable problem
This is where the conversation gets ugly, and it’s the point I care about most, because it isn’t cosmetic — it’s structural.
Almost all of these platforms charge by credits or tokens. Every message you send to build or fix something burns a variable amount based on complexity. It sounds reasonable until you remember one detail: these tools also introduce bugs. And when you ask the tool to fix the bug it just introduced, that fix costs too.
Read that again. You pay to correct the mistakes the tool itself made. The business model literally monetizes its own unreliability.
This isn’t theoretical. When v0 moved to token-based billing in May 2025, the community erupted. Under the old system, fixing a bug v0 had introduced was just another message; under credits, every attempt costs tokens. One user put it perfectly: “fixing one mistake can take 3-4 prompts, and now each of those costs me more tokens. It feels like I’m getting penalized for the product’s shortcomings.” There are reports of people spending $30 in a single day to add a couple of pages, with part of the spend going to fix what v0 itself broke. Vercel said, on the record, that they will not be reverting the change — and around February 2026 it roughly doubled per-token rates while leaving the seat price unchanged, which many read as a hidden price increase.
At Lovable the dynamic is identical: there are users reporting 400 credits burned in under an hour fixing bugs the AI introduced. The incentive is bent from the factory: the platform makes money when it gets things wrong. No healthy company designs that on purpose, but consumption-based billing produces it all on its own, and nobody’s in a hurry to fix it.
the number nobody publishes: churn
If you want to know whether a subscription business is healthy, don’t look at ARR. Look at retention. And here a very telling silence sets in.
Lovable has declined to share its churn rate or the split between monthly and annual subscribers. Churn — also called the cancellation or attrition rate — is simply the percentage of customers who drop their subscription in a given period; it’s the flip side of retention, and in a subscription business it’s the vital sign that actually matters. When a company proudly publishes every ARR milestone but hides retention, it’s usually because retention is the part it doesn’t want you looking at. Bolt’s own CEO said it without anesthesia: “the churn rate for everyone is really high. You have to build a retentive business” — speaking of the whole category, not a competitor. Bolt changed its subscription model precisely to stop the bleeding.
And there’s a public signal to match: traffic to Lovable fell around 40% from its peak. It’s not that nobody arrives — it’s that plenty of people arrive, try it, build their prototype, and leave. On top of that, every app a user builds costs the platform real money, because it pays Anthropic or OpenAI per inference. Record ARR + sky-high churn + a real marginal cost per user is a combination that looks gorgeous in a headline and a lot less gorgeous in a spreadsheet.
This doesn’t make Lovable a fraud — it’s a real company with a real product. But it does explain why the $6.6B valuation rests on an assumption, not a fact: that those millions of users who walk in will stay. The numbers they do publish don’t prove it, and the one that would is precisely the one they don’t publish.
and now their own supplier competes with them: Claude Design
There’s a twist that makes everything above more delicate. In April 2026 Anthropic launched Claude Design, its own tool for generating prototypes, landing pages and screens from a prompt — and it didn’t sell it separately: it folded it into the Claude Pro and Max subscription that millions of people already pay for.
Read it from Lovable’s or Bolt’s side. These platforms pay Anthropic (or OpenAI) per inference: the model that makes them work belongs to their new competitor. And for anyone already paying for Claude, Claude Design removes the need to pay for a separate tool for the most common use case — the quick prototype, exactly the 80% where these tools shine. On announcement day, Figma’s stock fell around 7%; the message to the market was clear.
Let me be honest about what I don’t know: there are no public figures isolating exactly how many users Lovable or v0 lost to Claude Design, and I’d rather not invent them. But the direction of the wind isn’t ambiguous. When your model provider becomes your competitor and gives away your flagship use case inside a subscription your customer already holds, the pressure on retention — that churn nobody publishes — only moves one way.
the part that doesn’t fit in the valuation: security
This is the section where I have to ask you not to read it as sensationalism, because the facts alone already sting enough.
In January 2025, CVE-2025-48757 was documented: a Row Level Security misconfiguration in Lovable-generated apps left more than 170 applications exposing emails, API keys, payment data, and personal information to anyone, with no authentication. The root cause was that the generated code skipped the access checks it was supposed to apply.
It wasn’t a one-off. In February 2026, an incident exposed 18,697 records at universities like UC Berkeley and UC Davis. And in April 2026, a researcher disclosed a mass leak affecting every project created before November 2025: just by creating a free account, anyone could read another tenant’s source code. Three serious incidents in thirteen months.
The pattern matters more than any individual breach. The promise of these tools is “you don’t need to know how to code.” But security — RLS, secret handling, input validation — is exactly the kind of thing that doesn’t show up in the demo and that you only know is missing once it has already failed. One security firm tested five popular vibe coding tools and found 69 vulnerabilities, six critical. That’s not bad luck: it’s what happens when you generate the visible part of the software and let the user assume the invisible part came included.
the 80/20 wall
And here, to me, is the heart of it. These tools are genuinely good — sometimes astonishing — at the first 80% of a project: the mockup, the screens, the happy path, the demo that makes your client say “yes, that.” The problem is the other 20%: the edge cases, the integrations, the production hardening, the security. And that 20% is exactly where projects die — because it requires precisely the engineering skills the tool promised you wouldn’t need.
The ecosystem data backs the feeling. Developer trust in AI-generated code dropped from 77% to 60%; nearly half of generated code samples are reported to contain OWASP Top-10 vulnerabilities; code churn and duplication went up. Generating code was never the bottleneck. Maintaining it is.
so, are they worth it?
Yes. But not for what the marketing sold you.
They’re absolutely worth it as a prototyping tool. To validate an idea in an afternoon, take a sketch into a meeting, show a client something tangible instead of a dead Figma, or kick off a throwaway project — they’re the best that has ever existed. There the “first 80%” is all you need, and they deliver it blazingly fast.
They’re not worth it as the factory for your real product. The moment that prototype has to withstand real users, real data, and maintenance under pressure, the model breaks: you pay to fix its own bugs, you inherit security decisions you didn’t make, and you’re left with code that different models wrote on different days with different judgment, with no owner and no shared style.
That’s why the workflow becoming consensus among people who do this seriously is hybrid: prototype in Lovable, Bolt or v0; graduate the real logic to Cursor or Claude Code once the idea proves it deserves to exist. It’s not a betrayal of any tool — it’s using each one for what it’s actually good at.
There’s a rule I apply to any new technology before I let it into a project: does it solve a problem I already have, and can it be maintained under pressure — with deadlines, bugs, and annoyed clients? These platforms pass the first question with honors and fail the second. And that’s perfectly fine, as long as you know which of the two you’re standing in.
they didn’t go out of fashion — they grew up
So when someone asks me again whether Lovable and v0 “went out of fashion,” I’ll give the same answer: no. What happened is less dramatic and more interesting. They stopped being the magic toy that was going to replace engineering teams and settled into their real place — an incredibly good prototyping layer, with a business model that rewards its own mistakes and a security record that doesn’t yet deserve your blind trust.
The mistake was never using them. The mistake is believing engineering judgment came included in the subscription. That part is still yours — and, ironically, it’s worth more than ever.