Guide · from the data

The vibe coding stack: what 1,503 products built with Claude Code, Cursor and Lovable run on

Of the 1,503 makers' products we track that say they're built with an AI coding agent or app builder, the most common stack is Next.js, Vercel, Supabase, Clerk, Stripe, Resend, OpenAI API and PostHog. It barely changes from one AI tool to the next, and most of what sets it apart from older products comes from being new, not from the AI: against products launched in the same years, the AI-built ones lean only towards Supabase for database (70% against 54%), Resend for transactional email (79% against 71%) and PostHog for product analytics (88% against 81%).

Published
October 20, 2026
Numbers refreshed
November 3, 2026
Reading time
4 min
Share
Post on X

We track 1,503 makers' products whose makers say they were built with an AI coding agent or app builder: Claude Code (772), Cursor (542), OpenAI Codex (145), Lovable (75), OpenCode (44), v0 (41), Bubble (38) and Devin Desktop (formerly Windsurf) (37). Each comes with a source: the maker's own words on Product Hunt or in an interview, or the tool's own showcase. What they chose for the rest of their stack comes from the same kinds of sources, the product's live site, DNS and code included.

The short answer

The most common pick at each step, among all of them:

  • Next.js for frontend framework: 284 of the 350 we know the frontend framework of (81%).
  • Vercel for hosting: 409 of the 672 we know the hosting of (61%).
  • Supabase for database: 188 of the 290 we know the database of (65%).
  • Clerk for authentication: 53 of the 72 we know the authentication of (74%).
  • Stripe for payments: 159 of the 221 we know the payments of (72%).
  • Resend for transactional email: 232 of the 345 we know the transactional email of (67%).
  • OpenAI API for LLM API: 180 of the 238 we know the LLM API of (76%).
  • PostHog for product analytics: 95 of the 119 we know the product analytics of (80%).

That's the default for a reason that matters more with an AI agent writing the code: these are the most documented choices, so the agent has seen the most examples of them.

It barely depends on the tool

The top pick, and its share, among the products built with each of the 5 most used AI tools; the last column is every maker's product, AI-built or not.

Claude CodeCursorOpenAI CodexLovableOpenCodeAll makers
Frontend FrameworkNext.js 78%Next.js 81%Next.js 84%Next.js 85%Next.js 69%Next.js 76%
HostingVercel 59%Vercel 66%Vercel 59%Vercel 53%Vercel 58%Vercel 57%
DatabaseSupabase 68%Supabase 61%Supabase 76%Supabase 94%Supabase 50%Supabase 45%
AuthenticationClerk 78%Clerk 67%Clerk 64%——Clerk 80%
PaymentsStripe 66%Stripe 76%Stripe 59%Stripe 87%Stripe 75%Stripe 77%
Transactional EmailResend 76%Resend 59%Resend 84%Resend 42%Resend 80%SendGrid 46%
LLM APIOpenAI 74%OpenAI 74%OpenAI 74%OpenAI 88%—OpenAI 75%
Product AnalyticsPostHog 87%PostHog 77%PostHog 89%PostHog 83%—PostHog 65%

A dash: fewer than five of that tool's products where we know the decision. Shares are of the products where we know it, so each cell has its own sample.

Is it the AI, or just new?

AI-built products are mostly new: 1,200 of the 1,503 launched in 2025 and 2026. New products already choose differently from old ones (most of the switch from SendGrid to Resend happened between launch years, AI or not), so comparing them with every product overstates what the AI tool changes. Held to the same launch years, 1,200 AI-built products against 2,104 whose makers named no AI coding tool:

PickAI-builtThe rest
Frontend FrameworkNext.js82% of 25877% of 640
HostingVercel60% of 51357% of 976
DatabaseSupabase70% of 23354% of 487
AuthenticationWorkOS14% of 493% of 154
PaymentsStripe69% of 15065% of 311
Transactional EmailResend79% of 23671% of 452
LLM APIGemini API17% of 15511% of 399
Product AnalyticsPostHog88% of 7781% of 155

The AI-built ones lean towards Supabase for database (70% against 54%), Resend for transactional email (79% against 71%), PostHog for product analytics (88% against 81%), Gemini API for LLM API (17% against 11%) and Next.js for frontend framework (82% against 77%). For hosting and payments, they choose like everyone else launching now. "The rest" doesn't mean no AI: those makers simply didn't name a coding tool, and many surely use one.

Tool by tool

Claude Code, 772 products: frontend framework: Next.js (135 of 172); hosting: Vercel (203 of 345); database: Supabase (101 of 149); authentication: Clerk (29 of 37); payments: Stripe (71 of 108); transactional email: Resend (122 of 160); LLM API: OpenAI API (71 of 96); product analytics: PostHog (40 of 46).

Cursor, 542 products: frontend framework: Next.js (109 of 135); hosting: Vercel (176 of 266); database: Supabase (71 of 117); authentication: Clerk (20 of 30); payments: Stripe (68 of 90); transactional email: Resend (83 of 141); LLM API: OpenAI API (87 of 118); product analytics: PostHog (46 of 60).

OpenAI Codex, 145 products: frontend framework: Next.js (37 of 44); hosting: Vercel (41 of 70); database: Supabase (13 of 17); authentication: Clerk (7 of 11); payments: Stripe (10 of 17); transactional email: Resend (27 of 32); LLM API: OpenAI API (14 of 19); product analytics: PostHog (8 of 9).

Lovable, 75 products: frontend framework: Next.js (11 of 13); hosting: Vercel (16 of 30); database: Supabase (17 of 18); payments: Stripe (13 of 15); transactional email: Resend (8 of 19); LLM API: OpenAI API (14 of 16); product analytics: PostHog (5 of 6).

OpenCode, 44 products: frontend framework: Next.js (11 of 16); hosting: Vercel (14 of 24); database: Supabase (3 of 6); payments: Stripe (6 of 8); transactional email: Resend (12 of 15).

Smaller samples are smaller for a reason: app builders host more of the product themselves, so fewer of its choices show from outside.

If you're starting a vibe-coded app

The default above is a safe start because it's what agents know best, not because it's right for every product. A product that sells to companies may need SSO-ready auth; one with long-running jobs needs a host that keeps a process running. Plan a stack for your situation — it asks six questions and prices the result — and give your agent an AGENTS.md written for your framework.

Counted from makers' products with a source for the AI tool and for each decision. A tool used only in development rarely shows from outside, so these products are the ones whose makers said so. How evidence is collected.

Questions

What is the best tech stack for vibe coding?

The one most makers' AI-built products run on: Next.js for frontend framework (81% of 350), Vercel for hosting (61% of 672), Supabase for database (65% of 290), Clerk for authentication (74% of 72), Stripe for payments (72% of 221), Resend for transactional email (67% of 345), OpenAI API for LLM API (76% of 238) and PostHog for product analytics (80% of 119). These are the best documented picks, which is what an AI agent writes best. Whether they fit your product is a separate question: plan one for your situation at howmakersbuild.com/plan.

What tech stack do apps built with Claude Code use?

Across 772 makers' products built with Claude Code: frontend framework: Next.js (135 of 172); hosting: Vercel (203 of 345); database: Supabase (101 of 149); authentication: Clerk (29 of 37); payments: Stripe (71 of 108); transactional email: Resend (122 of 160); LLM API: OpenAI API (71 of 96); product analytics: PostHog (40 of 46).

What tech stack do Lovable apps use?

Across 75 makers' products built with it: frontend framework: Next.js (11 of 13); hosting: Vercel (16 of 30); database: Supabase (17 of 18); payments: Stripe (13 of 15); transactional email: Resend (8 of 19); LLM API: OpenAI API (14 of 16); product analytics: PostHog (5 of 6).

Does the AI coding tool change the stack?

Little. Against products launched in 2025 and 2026 whose makers named no AI coding tool, the AI-built ones chose Supabase for database (70% against 54%), Resend for transactional email (79% against 71%), PostHog for product analytics (88% against 81%), Gemini API for LLM API (17% against 11%) and Next.js for frontend framework (82% against 77%); for hosting and payments they choose the same.