Non-technical founder — ship with AI

Who this is for

Founders with an idea and domain knowledge but no coding background: the guests here include an ex-optician, a diplomat, a finance graduate, a quant trader and an Amazon seller. Almost all the AI-building evidence is from 2025–26, so the tool names will date quickly. Treat the build times and revenue figures as self-reported.

The route at a glance

1. Learn by building one real project.
2. Design first, from apps that already work.
3. Plan with AI, give it context, and build in micro-steps.
4. Prototype in an app builder, finish in Cursor or Claude Code, and paste errors back.
5. Expect hallucinations, rework and some paid help.
6. Or bring in a technical partner with skin in the game.
7. Lean into your non-technical edge, because marketing is now the bottleneck.

Step 1 — Learn by building a real project

Pick something you actually want to ship and learn only what it needs. Samuel Rondot, a former optician with “zero coding experience”, followed a 15-hour YouTube course while rebuilding an Instagram tool, applying each lesson straight away. He says watching a course on its own is not enough (Samuel Rondot · Samuel Rondot (Jul 2025) ▶ 01:19). His advice today is to skip traditional coding: pick a real project, ask ChatGPT what you need to learn, and ask again at every roadblock (Samuel Rondot · Samuel Rondot (Jul 2025) ▶ 02:03). Guests report getting productive in about two months (Denis Yurchak (Denis Yurchak · Denis Yurchak (Oct 2025) ▶ 03:30); Nico, who locked himself in a hotel room (Nico · Nico (Mar 2025) ▶ 04:05)). The route can be scrappy: Sam Adeniyi pasted AI-written code into Notepad until a friend told him about Cursor, which he took to mean a mouse cursor (Sam Adeniyi · Sam Adeniyi (Jan 2026) ▶ 04:11).

Common mistake: a first project that is too big. Ken built a small card game first so he could learn AI-assisted development and app-store approval from start to finish (Ken · Ken (Sep 2026) ▶ 04:14). Connor Burd spent six to eight months on a social app that failed before switching to utilities (Connor Burd · Connor Burd (Oct 2025) ▶ 02:26).

Step 2 — Design first, from reference apps

Before any code, collect screens from apps that already work (Design first from reference apps, then build with AI). Connor Burd starts by downloading about 20 apps in his category (Connor Burd · Connor Burd (Oct 2025) ▶ 03:16). Erikas Malisauskas notes what he likes and dislikes about competitors’ UX so he can design something better (Erikas Malisauskas · Erikas Malisauskas (Apr 2025) ▶ 08:44). Ania Wysocka sketched in a notebook, taught herself wireframing in Photoshop and Illustrator, and prepared all the content before development began (Ania Wysocka · Ania Wysocka (Aug 2025) ▶ 03:48). Benji works back from the value proposition the app has to show (Benji · Benji (May 2026) ▶ 03:50). Note that Mobbin, a library of app screens, sponsored episodes on the channel, including Samuel Rondot’s.

Step 3 — Plan with AI, give it context, build in micro-steps

Let an AI break the project into steps, then build one small block at a time (Plan with AI, then build in micro-steps). Paulius Masalskas tells Perplexity he is “a complete noob” to get a step-by-step reference document (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 03:55). Alex Finn asks ChatGPT for micro-steps and feeds them to Cursor one at a time, saying the smaller the request, the fewer the bugs (Alex Finn · Alex Finn (Jun 2025) ▶ 04:25). Mau Baron builds feature by feature, because AI cannot yet make a whole app from one prompt (Mau Baron · Mau Baron (Jan 2026) ▶ 05:37) (Mau Baron · Mau Baron (Jan 2026) ▶ 05:28).

Give the model context so it does not have to guess. Connor Burd writes a data spec with JSON examples and drops design screenshots into Claude or Cursor (Connor Burd · Connor Burd (Oct 2025) ▶ 05:07) (Connor Burd · Connor Burd (Oct 2025) ▶ 05:31). David Attias screenshots every screen of the app he is cloning and asks Cursor to rebuild each one (David Attias · David Attias (Oct 2025) ▶ 05:20). Max asks for a detailed implementation plan with UX constraints (Max · Max (Dec 2025) ▶ 06:40).

(CJ Zafir · CJ Zafir (May 2025) ▶ 17:33)

Step 4 — Prototype in a builder, finish in an IDE, paste errors back

A pipeline that recurs is an AI app builder (Bolt, Lovable, Replit) for the first version, then Cursor or Claude Code to finish it. Paulius Masalskas says Bolt gets you “80% of the way there” and Cursor takes it over the line (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 04:59); Presh uses Lovable, Replit or Bolt to test ideas quickly (Presh · Presh (Dec 2025) ▶ 08:47). When something breaks, paste the error back. Paulius does this in Bolt (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 04:22); George Lampropolis pasted Expo logs into ChatGPT (George Lampropolis · George Lampropolis (Dec 2025) ▶ 04:25). Jack Friks learned to code by feeding every error to ChatGPT until the app worked (Jack Friks · Jack Friks (Mar 2025) ▶ 03:22).

(Cedric Roberge · Cedric Roberge (May 2026) ▶ 03:44)

Reported build times run from about three days of coding (Mau Baron (Mau Baron · Mau Baron (Jan 2026) ▶ 04:01)) to two weeks (Tomer (Tomer · Tomer (Aug 2026) ▶ 03:21); Cedric Roberge (Cedric Roberge · Cedric Roberge (May 2026) ▶ 03:47)), plus a week or more for App Store approval. All of these are self-reported. Presenter Pat Walls’s intro to Alex Finn’s episode says three months, where Alex himself describes seven or eight (presenter · Alex Finn (Jun 2025) ▶ 00:09).

Step 5 — Expect hallucinations, rework and some paid help

The same guests who say the barrier has gone also say the tools need supervision. CJ Zafir says the models “hallucinate, they assume things” (CJ Zafir · CJ Zafir (May 2025) ▶ 02:14); the tool he built, CodeGuide, sells a fix for exactly that problem (CJ Zafir · CJ Zafir (May 2025) ▶ 02:21), which is a commercial interest. Denis Yurchak had to correct the design by hand (Denis Yurchak · Denis Yurchak (Oct 2025) ▶ 05:34). Jack, an e-commerce consultant, spent about three and a half months correcting what Cursor got wrong in his spreadsheet-to-app build (Jack · Jack (Jun 2026) ▶ 06:35).

(presenter · Steph France (Jun 2025) ▶ 23:07)

That line is Pat’s, as presenter, and he warns elsewhere that people who try to “one-shot” an app often get lost and give up (presenter · Connor Burd (Oct 2025) ▶ 06:12). Budget for help on the hard parts: George Lampropolis paid a Fiverr developer about $250 for payments and authentication (George Lampropolis · George Lampropolis (Dec 2025) ▶ 04:35). He also says vibe coding is for validating an MVP, and that past about 5K a month (currency unstated) you should invest in product quality (George Lampropolis · George Lampropolis (Dec 2025) ▶ 10:33).

Step 6 — Or bring in a technical partner with skin in the game

The older route still works. Colin McIntosh says you will never convince a random engineer to build your idea without paying them a lot, “nor should they” (Colin McIntosh · Colin McIntosh (Jan 2025) ▶ 08:11). He found partners through long-standing relationships and split the equity three ways (Colin McIntosh · Colin McIntosh (Jan 2025) ▶ 08:38). Ben Leavitt says to pay for a great developer and offer skin in the game, such as continued employment or a revenue split (Ben Leavitt · Ben Leavitt (Mar 2026) ▶ 14:55) (Ben Leavitt · Ben Leavitt (Mar 2026) ▶ 14:25). Erikas Malisauskas found his technical co-founder by posting in a local developer Facebook group, and treated it like sales (Erikas Malisauskas · Erikas Malisauskas (Apr 2025) ▶ 09:38). Steven Cravotta hires Upwork developers and pays per completed project, not by the hour (Steven Cravotta · Steven Cravotta (Sep 2024) ▶ 06:02).

Step 7 — Lean into your non-technical edge

Once building is easy, distribution is the constraint. Developers admit it:

(Christian Konnerth · Christian Konnerth (Sep 2025) ▶ 08:18)

Jure Sotosek says many programmers give up “when there comes time to sell” (Jure Sotosek · Jure Sotosek (Apr 2025) ▶ 03:01). Roman thought being non-technical was a disadvantage but now calls it his “secret sauce”, because of his interest in psychology (Roman · Roman (Sep 2026) ▶ 02:59). Sarah Pearl argues that women “dominate distribution” and that coding was the only barrier (Sarah Pearl · Sarah Pearl (Jul 2026) ▶ 07:39). Ethan used AI to shift his time to learning sales and marketing (Ethan · Ethan (Apr 2026) ▶ 03:12).

Where guests disagree

• Do you still need a technical co-founder? Nick Sweeney says ideally yes (Nick Sweeney · Jack Sweeney & Nick Sweeney (Nov 2025) ▶ 07:04). Paulius Masalskas calls “AI as your CTO” a superpower (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 12:34), and Hassam says you no longer need to be an engineer (Hassam · Hassam (Jan 2026) ▶ 10:49).
• How big is the gain? Nico reports working about ten times faster (Nico · Nico (Mar 2025) ▶ 13:52). Marc Lou says people “fall for the productivity trick” (Marc Lou · Marc Lou (Apr 2026) ▶ 07:40).

What can go wrong

• “Non-technical” stories often include hired developers: Pat points this out about one guest (presenter · David (May 2026) ▶ 05:39), and Ben Leavitt says “I’m not the developer” (Ben Leavitt · Ben Leavitt (Mar 2026) ▶ 13:53).
• AI makes it easy to skip talking to users. An unidentified presenter (Pat or Gus) says the tools make it “so easy to… avoid the hard part” (unnamed guest · Brian Shin (May 2026) ▶ 14:23).
• The window may close: Paulius does not know how long it will last (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 12:37).
• Possible paid placements and self-interest. Bubble sponsored the Bubble-built no-code episode (x9TUDb4sLE0), so guest praise of Bubble may be paid placement. RevenueCat hosted the hackathon Connor Burd won and is in his stack. Ads from AI app builders (Rocket, Emergent, 10Web) were removed from some episodes. The channel also plugs its own AI-building course. The presenters’ enthusiasm for AI building should be read with that in mind.

How strong is the evidence

Medium. Steps 1, 3 and 7 each have six or more independent guests. Step 2 rests on about five, and step 4 on four or five. The partner route (step 6) is well supported but older. Pat’s many AI-building lines count as the presenters’ view, not guest evidence. The titles’ revenue figures and the “didn’t write a line of code” claims are self-reported, and several differ from what was said on camera.

Go deeper

Building as a non-technical founder · Vibe coding · No-code builders · Co-founders & partners · Contractors & offshore talent · Tech stack · Plan with AI, then build in micro-steps · Design first from reference apps, then build with AI · How did you build it and how long did it take? · What’s your tech stack (and what does it cost)? · Founder operating system — routines, deep work and avoiding burnout