Subscribe
Analysis
6 min read

The Prompt Factory: What Lovable, Replit, Bolt, v0 and Base44 Actually Sell in Late 2026

The vibe-coding platforms now post nine-figure run rates, charge for a usage meter rather than software, and share one well-documented security defect: open databases.

Mid-century industrial-design poster of a control console that turns a written sentence into small buildings and app windows
Mid-century industrial-design poster of a control console that turns a written sentence into small buildings and app windows

The prompt-to-app business is a little over two years old and now has revenue that most software companies twice its age would envy. It also has a shared security problem that no vendor is fixing on its own. This is what the five major platforms actually do differently, what they cost, and where the model breaks.

Five prompt-to-app platforms as workshop machines in a row

The numbers first

The category's financials stopped being speculative sometime in 2025 and got weird in 2026.

Lovable, the Stockholm company that launched publicly in November 2024, reported $200M ARR at its one-year mark, took a $330M Series B at a $6.6B valuation in December 2025, and raised a $400M Series C led by Menlo Ventures and EQT's Scaleup Europe Fund in August 2026 at $13.3B, with ARR tracking toward $600M around the time of the round. Users have created more than 60M projects on the platform, and Lovable says apps built there draw over 900M visits per month. Roughly two-thirds of Fortune 500 companies have employees using it, up from half six months earlier.

Replit, the nine-year-old browser IDE that pivoted to agents, raised $250M at a $3B valuation in September 2025 when it was on track for $150M in annualized revenue, then tripled that valuation in March 2026 with a $400M Series D led by Georgian at $9B. The company says it expects to cross $1B in run-rate revenue by the end of 2026 and that users come from 85% of the Fortune 500.

Bolt.new (StackBlitz) is the cautionary tale in the group. It went from $0 to about $40M ARR in five months after its October 2024 launch, reportedly one of the fastest-growing software products ever at the time, and StackBlitz raised a $105.5M Series B at a roughly $700M valuation in January 2025. Public numbers have been quieter since; the $40M figure from March 2025 remains the last widely cited one.

Base44 is the one that already exited. Wix acquired the six-month-old, solo-founded company in June 2025 for roughly $80M in cash plus earn-outs through 2029. Base44 reportedly crossed $100M ARR by early 2026 and $150M by May 2026, and Wix's Q2 2026 results disclosed an additional $41M earn-out payment, putting total consideration past $150M for what was eight employees at signing.

v0 is Vercel's wedge in the same market: it runs $30 per user per month on its paid tiers, and Vercel overall was valued at $9.3B in its September 2025 Series F.

What each one actually sells

Strip the marketing and the five platforms sell four different things.

v0 sells components. It is the best at producing polished React/Next.js with shadcn UI that drops into an existing codebase, and its output is a frontend layer, not an operating business. Reviewers consistently classify it as a UI/component generator rather than an app builder. Fine if your team already runs the rest.

Bolt sells the browser as a computer. Its differentiator is StackBlitz's WebContainer runtime: a real Node environment, terminal, packages and preview, all inside a tab, no local setup. It builds full-stack apps and added a prompt-to-mobile path through an Expo partnership in February 2025. The trade-off is depth: its backend story is narrower than the others.

Lovable and Base44 sell the business. Both target non-technical founders and handle auth, database, hosting and payments in one bill. Lovable's default stack is React/TypeScript/Tailwind with Supabase on the backend; Base44 is the most vertically integrated, with storage, email, analytics and auth all built in so there is no stack to assemble. Both export code to GitHub, which is the most important checkbox in the category.

Replit sells the platform. Agent plus a full IDE, multi-language backends (Python, Node, Go and more), a 200-minute autonomous runtime in Agent 3, Design Mode mockups, deployment and distribution through Replit Apps. It asks more of the user than Lovable does and gives more rope.

The practical fork that even the vendors' own comparison pages concede: if you want a demo today, Bolt; if you need Python or persistent servers, Replit; if you want production-looking web apps described in plain language with auth and payments, Lovable or Base44; if you are a React team, v0.

The pricing model: nobody sells software, everyone sells a meter

Every platform starts free or cheap and bills usage.

PlatformEntry paid planMeter
LovablePro $25/mo, Business $50/moCredits; monthly credits expire two months after issuance; hosting metered on top
ReplitCore $20/moModel-usage credits layered on subscription; pay-as-you-go when they run out
BoltPro $25/moTokens per generation and fix
v0$30/user/moPer-model token pricing plus small monthly and daily credit grants
Base44Starter ~$16/mo billed annuallyCredits per generation

Two consequences, per the vendors' own terms. First, real project cost runs roughly double the sticker price because every correction to generated code consumes the meter again. Second, this is why ARR grows faster than seat counts: heavy users pay far more than the headline plan, and hosting is metered separately. The revenue figures above are as much a story about usage billing as about adoption.

The shared debt

The security picture is the part the funding announcements skip.

A vault door standing ajar with data spilling out

A June 2026 academic study ("Understanding the (In)Security of Vibe-Coded Applications," arXiv:2606.23130) collected 9,041 open-source applications built with AI agents (Claude Code and Lovable named explicitly) and manually audited 200 publicly deployed ones. Result: 1,186 validated vulnerabilities, 91% of audited apps containing at least one, and 65.8% of the vulnerabilities rated Critical or High, concentrated in broken access control, injection and authentication failures. The authors trace them to eight recurring failure modes rooted in systematic agent limits: memory, objective and knowledge defects.

The most expensive single failure mode is database configuration. Vibe platforms make it trivial to bolt a Supabase Postgres backend onto an app, and the agents often create tables programmatically without enabling row-level security. UpGuard found more than 16,000 publicly exposed Supabase databases, over half holding personal data such as names, phone numbers and passwords. The misconfiguration tied to Lovable-generated apps was tracked as CVE-2025-48757 as early as March 2025; Supabase changed its Table Editor to enable RLS by default, but tables created through the API, which is how coding agents work, still do not get it by default.

Individual incidents fill in the rest. Moltbook, an AI social network its founder said was built entirely with AI coding tools, leaked 1.5M API tokens and tens of thousands of emails within days of launch because a Supabase key sat in client-side JavaScript with no RLS behind it (Wiz). Escape.tech scanned 5,600 vibe-coded apps and found over 2,000 exploitable vulnerabilities and 400+ exposed secrets. Red Access swept 380,000 URLs and found 2,000 of 5,000 accessible AI-built apps exposing sensitive data. And IBM's analysis of GitGuardian data notes that secrets exposed in AI-assisted commits are less likely to be revoked than human-exposed ones: 64% of credentials validated in 2022 were still exposed as of January 2026.

The uncomfortable inference for the platform vendors: their flagship backend integration is the leading cause of their users' breaches, and their target customer is precisely the person who will not audit the RLS policy. Lovable and others now ship security guidance and default hardening, but the academic evidence says insecurity is still the norm in deployed apps, not the exception.

What to conclude

For prototypes, internal tools and demos, the category works, the prices are near-identical, and the differences are about where the output lives (components vs. owned code vs. platform runtime) rather than whether it works. Choose by ecosystem, and export to GitHub on day one regardless of vendor.

For anything touching customer data, the burden of proof has not moved: a 200-app audit put a critical-or-high vulnerability in most deployed vibe-coded apps, and the single most common defect, an open database, is invisible to the person buying the platform. The revenue is real. The governance around it is not yet.

Sources

AnalysisTool Comparison