How do founders build apps fast with AI coding tools?

By gr0w Research · Updated · From 185 founder interviews · 6 min read

Short answer

In an analysis of 185 founder interviews, founders who built with AI tools reported first versions in anywhere from 3 days (Prayer Lock, built with Cursor and ChatGPT) to about 8 weeks (Magai, built on Bubble). They built one core feature first, fed the AI small steps, and skipped extras like onboarding or login until after the first version worked. Tool choice varied (Cursor, Claude, Bolt, Rork, Bubble), and the founders who said what came next put the time they saved into distribution.

3 days
time Mao Baron said he spent coding Prayer Lock with Cursor and ChatGPT
under 2 weeks
time Connor said he spent vibe coding Payout with Claude
8 weeks
time Dustin said he spent building Magai's MVP on Bubble, with no developer
185
founder interviews searched for this page

Which AI coding tools did founders actually use?

Founders named a mix of code editors, chat assistants and app builders, and most combined two or three. The library does not compare tools head to head, so what follows is what each founder said they used.

  • Cursor: Mao Baron built Prayer Lock with Cursor on Sonnet daily, Codex CLI for complex tasks and ChatGPT for questions, partly to keep token costs low [1]. Tomer vibe-coded Cal Buddy with Cursor without a developer background [2].
  • Claude: Connor said Claude is the only AI coding tool he uses now [3]. In another interview he said he was undecided between Claude Code and Cursor and mostly picked Cursor [4].
  • Bolt then Cursor: Paulius built Creator Hunter's MVP in Bolt, then downloaded it into Cursor to add a Supabase backend and Clerk auth [5].
  • Rork, Lovable, Replit: George built Wrestle.ai in Rork and called it the best vibe-coding tool he tried, at $25 a month [6]. Lino Leiden's first Lovable version was too poor to release, so he rebuilt in Rork [7].
  • Switching: Alex Finn said Cursor's quality dropped around October 2024, so he moved to Windsurf [8].

How long does it take to build an MVP with AI?

Founders reported anywhere from two hours to about two months, and the fastest were simple, single-feature apps. Here is what they said:

  • Max, an iOS engineer, said his fastest app took 2 hours, because he reused onboarding, paywall and settings code from earlier apps, sometimes 90% of it [9].
  • Mao Baron coded Prayer Lock in 3 days, then waited over a week for App Store approval [1].
  • Connor built Payout in under two weeks with Claude [3]. Tomer built Cal Buddy in two weeks, and said Apple approval and review took longer than coding [2].
  • George went from idea to the App Store in about a month with Rork and ChatGPT [6].
  • Dustin took 8 weeks on Bubble [10].

Speed was not guaranteed. Lino Leiden spent two months prompting his first Lovable version before rebuilding [7]. Alex Finn had a prototype on day one but spent months iterating with beta testers before launch [8].

What should you build first?

Founders built one core feature first and left everything else for later. Max studies two or three competitors and copies only the single core feature tied to the keyword he is targeting [9]. Connor builds core functionality first, skips onboarding at first, and drops screenshots into Claude to generate full screens [3]. He also said ship a version fast rather than being picky [3].

Mao Baron went further: no authentication and no database, with everything stored on the device, to cut signup friction and costs [1]. Jacob and Alex's first month on Faceless Video was one technical proof only: type text, get a faceless video. Users, payments and pricing came afterward [11].

Several founders chose proven ideas over new ones. Mao Baron improved on an existing app after about three months of building apps nobody wanted [1]. Tomer localized Cal AI into Hebrew for a market that didn't know it [2]. Connor later returned to onboarding, saying about 90% of users in a paywalled app only see that [3].

How do you prompt AI so you don't get stuck?

Founders broke work into small steps, gave the AI written context, and pasted errors straight back in. Alex Finn split every feature into the smallest pieces (input field, then button, then AI call) and had ChatGPT write those micro steps to paste into Cursor [8] [8]. Paulius used Perplexity deep research to get a step-by-step document, built one step at a time in Bolt, and threw errors back at the AI [5] [5].

Connor wrote a text document of his data structures with sample JSON so the AI wouldn't guess [3]. Max asked ChatGPT, Gemini or Claude for a plan listing every screen and feature before coding [9]. For design, Connor and Mao Baron collected screenshots of other apps' onboarding into Figma or Canva and built from the best ones [3] [1].

Connor also warned that people who try to one-shot something real often get lost, can't deploy, and give up [3]. Mao Baron likewise built feature by feature rather than in one shot [1].

Is no-code still a way to build fast?

Yes, several founders reached real revenue on Bubble without writing code. Dustin, who is not a developer, took an online course and had a full Magai MVP in 8 weeks on Bubble plus a little code he wrote himself [10]. Jacob and Alex built Faceless Video on Bubble plus APIs, and said Bubble handles security, scalability and backend [11] [11]. David Bressler built Formula Bot 100% in Bubble from free YouTube tutorials, and only moved 5–10% to code after limitations appeared more than a year later [12].

These interviews are mostly web apps. For mobile, Rork, Lovable and Bolt came up. Presh suggested testing an idea in Lovable, Replit or Bolt before building the native app [13]. George still paid a Fiverr developer about $250 to add payments and auth to his Rork app [6].

What do founders do once the app is built?

The founders who said what came next put the saved time into distribution. Mao Baron capped the build at two weeks, then stopped touching the app to focus on distribution [1]. Alex Finn said product is no longer the moat and distribution is [8]. Lino Leiden said to obsess over the product only once marketing is stable at $5–10K a month [7].

Speed also makes copying easy. Paulius said copycats of Creator Hunter never took off without an edge AI tools can't easily copy [5]. Peter of Stronger said a workout tracker can now be vibe-coded in two or three days [14].

Tools and time to first version, as founders said it (self-reported)
Founder / productToolsTime to first version
Mao Baron, Prayer LockCursor, ChatGPT, Codex CLI [1]3 days of coding, plus a week for App Store approval
Connor, PayoutClaude [3]Under 2 weeks [3]
Tomer, Cal BuddyCursor [2]2 weeks, then slower Apple review
George, Wrestle.aiRork, ChatGPT, $250 Fiverr dev [6]About 1 month, idea to App Store
Paulius, Creator HunterPerplexity, Bolt, Cursor, Supabase, Clerk [5]Not said
Dustin, MagaiBubble plus some code [10]8 weeks
Jacob and Alex, Faceless VideoBubble and APIs [11]First month was the technical proof only
Max, 28-app portfolioCursor, reused code [9]As little as 2 hours per app

Frequently asked questions

Can you build an app with AI if you can't code?

Several founders said they did. George had never written code and went from idea to the App Store in about a month with Rork and ChatGPT [6]. Dustin built Magai's MVP in 8 weeks on Bubble after an online course [10].

Cursor or Claude Code: which did founders prefer?

There was no consensus. Connor said Claude is his only AI coding tool [3], while in another interview he said he mostly picks Cursor and is undecided between the two [4]. Alex Finn moved from Cursor to Windsurf after feeling quality dropped [8].

What does an AI-built MVP cost to run?

Costs were small. George paid $25 a month for Rork, about $30 for Supabase and $40–60 for OpenAI [6]. Max said his tooling for 28 apps cost roughly $300 a month [9].

Should you add onboarding and login to the first version?

Founders mostly left them out first. Connor skipped onboarding initially [3], and Mao Baron used no authentication or database [1]. Connor added onboarding later, since about 90% of users in a paywalled app only see that [3].

What slows founders down after the code is done?

App Store review. Tomer said Apple approval took longer than coding [2], and Mao Baron waited over a week [1].

Sources

Founder interviews on YouTube; numbers are as the founders said them. How this research is done.

  1. [1]I Built a $20K/Month Mobile App: Here’s My Entire Playbook · Prayer Lock (Starter Story)
  2. [2]I Copied A Huge App. Now I Make $80K/Month · Cal Buddy (Starter Story)
  3. [3]I vibe coded a $20K/month mobile app in 14 days · Payout (Starter Story)
  4. [4]This 23 Yr Old Genius Makes $1M/Year with No-Code Apps. Here's How · Payout (Superwall)
  5. [5]I Built an App with Cursor, Made $30K, and Quit My Job. · Creator Hunter (Starter Story)
  6. [6]How I grew my mobile app to $17K per month · Wrestle.ai (Starter Story)
  7. [7]This crazy simple app makes $50K/month · AbMax (Starter Story)
  8. [8]I Built an App with Cursor and Made $100K on Launch Day · Creator Buddy (Starter Story)
  9. [9]I make $10K/month from 28 apps (Starter Story)
  10. [10]I built a $100K/month AI app · Magai (Starter Story)
  11. [11]I Built a $1M SaaS 100% with No Code (Bubble) · Faceless Video (faceless.video) (Starter Story)
  12. [12]I Built a $1M AI App [No Code] · Formula Bot (formerly Excel Formula Bot) (Starter Story)
  13. [13]I make $120K/year from 3 mobile apps · Go Polar, SunSeek, Posture AI, Tempo (Starter Story)
  14. [14]My App Makes $600k/Year Using This Easy Growth Strategy (copy me) · Stronger (Superwall)