AI product with a moat

Who this is for

You want to build a consumer AI app or an AI tool for a business workflow on top of models you do not own. This is one of the channel’s most covered topics: about 30 episodes discuss wrappers and moats, and about 32 each cover consumer AI apps and API costs.

It dates fast. Almost all the evidence is from 2024–26. Model prices, the capabilities that open new app ideas and the tools guests name change within months. Treat the tactics as a snapshot, not settled practice.

The route at a glance

1. Pick one workflow or one AI “magic moment”.
2. Go beyond a thin wrapper: own the whole workflow or build what the model owner can’t.
3. Treat brand, audience and distribution as the moat.
4. Monetise early and price around inference cost.
5. Hedge your dependence on platforms and models.

Step 1 — Pick a workflow or a single magic moment

Consumer AI apps on the channel usually centre on one moment: the user uploads something and gets an instant verdict or transformation. George Lampropolis calls his the “gotcha moment”: upload a wrestling match and get a score out of 10 with drills (George Lampropolis · George Lampropolis (Dec 2025) ▶ 15:23). Timo Koehler’s app analyses a photo of a stock chart and says it makes $56,000 a month behind a hard paywall (self-reported) (Timo Koehler · Timo Koehler (Jul 2026) ▶ 01:42). Other examples are makeup try-ons (Louis · Louis (Dec 2025) ▶ 01:33) and home restyling (Pauline · Pauline (Oct 2024) ▶ 01:23); food-photo calorie counters recur too.

Ideas often come from a new model capability or a trend. The Koehler brothers began when ChatGPT added image upload (Timo Koehler · Timo Koehler (Jul 2026) ▶ 03:45). Jozef Gherman built StealthGPT after TikTok showed him a video about an AI detector (Jozef Gherman · Jozef Gherman (Jul 2024) ▶ 06:23). Several founders say they sell an emotion rather than a utility; Kelechi says both his apps “really” sell confidence (Kelechi · Kelechi (Oct 2025) ▶ 11:14).

For business tools, Ramon Berrios (Castmagic) says the opportunity is to embed in an existing workflow and radically remove friction: a painkiller, not a vitamin (Ramon Berrios · Ramon Berrios & Blaine (Jun 2024) ▶ 01:38). Creator and marketing workflows are the biggest cluster, from scripting to scheduling posts (Gil Hildebrand · Gil Hildebrand (Dec 2025) ▶ 01:00). Another cluster is “picks and shovels” for AI builders: CJ Zafir says CodeGuide grew by borrowing the AI coding tools’ audience (CJ Zafir · CJ Zafir (May 2025) ▶ 02:21).

Common mistake: a generic AI toy. Piotr says he wanted a specific niche (Piotr · Piotr (Sep 2026) ▶ 03:53); his render tool sat at $100 a month for two years before a pivot took it to 10K (self-reported) (Piotr · Piotr (Sep 2026) ▶ 01:16).

Step 2 — Go beyond a thin wrapper

The most concrete warning comes from David Bressler (Formula Bot, 2024). Within weeks about ten near-identical Excel formula sites copied his product, and Microsoft sent a cease-and-desist before announcing its own OpenAI tie-up (David Bressler · David Bressler (Feb 2024) ▶ 11:49). Microsoft had also offered to build his tool into its add-on store, and he realised it could absorb the idea (David Bressler · David Bressler (Feb 2024) ▶ 13:13).

(David Bressler · David Bressler (Feb 2024) ▶ 12:29)

He says some of his churn still goes to free ChatGPT (David Bressler · David Bressler (Feb 2024) ▶ 14:11). Paulius Masalskas agrees you need an edge AI tools can’t easily copy (Paulius Masalskas · Paulius Masalskas (May 2025) ▶ 10:44).

The fix guests describe is product depth:

• Own the workflow. Berrios says ChatGPT solves one small pain point, not a workflow from A to Z (Ramon Berrios · Ramon Berrios & Blaine (Jun 2024) ▶ 08:13).
• Integrate across platforms. Zac’s Hiro connects Klaviyo’s rivals, which Klaviyo is unlikely to show (Zac · Zac (Jun 2026) ▶ 07:53).
• Hide the prompting. Piotr added a chat assistant that asks follow-up questions and writes the prompt (Piotr · Piotr (Sep 2026) ▶ 04:34).
• Ground answers in the user’s own data. Elephas answers only from documents the user adds (Ayush · Ayush (Jan 2026) ▶ 16:20).
• Radical simplicity. Anton says Letterly beats free alternatives because few products are truly simple (Anton · Anton (Nov 2025) ▶ 03:48).

Step 3 — Treat brand, audience and distribution as the moat

The most common answer on the channel is that the code is not the moat.

(Nevo · Nevo (Jan 2026) ▶ 09:26)

Alex Finn says a copycat with ten followers couldn’t outsell him with the identical app (Alex Finn · Alex Finn (Jun 2025) ▶ 12:03). Gil Hildebrand says people buy on belief in a brand or person (Gil Hildebrand · Gil Hildebrand (Dec 2025) ▶ 09:12). Breakout consumer apps usually grew through TikTok and creators. Natural Write reached $100K in three months from one TikTok account, according to the founders (Nikita · Nikita & Evgeniy (Sep 2025) ▶ 01:40). David Park’s Jenny AI ran one repeatable “essay due” format for months (David Park · David Park (Mar 2025) ▶ 21:24).

Read with care. Much of this framing is Pat’s, and he repeats it in several episodes: code has little value (presenter · Nevo (Jan 2026) ▶ 10:07), audience and distribution are what last (presenter · Rob Hallam (Dec 2025) ▶ 19:35), and content is how you stand out (presenter · Kyan (Jun 2026) ▶ 12:11). Label these as the presenters’ view. Several guests who make the argument also sell audience or content tools: Finn sells Creator Buddy (Alex Finn · Alex Finn (Jun 2025) ▶ 01:01), and Rob Hallam sells SuperX (Rob Hallam · Rob Hallam (Dec 2025) ▶ 02:09).

Step 4 — Monetise early and price around inference cost

AI APIs are usually the biggest running cost. Diego Roshardt reports about 2.5K a month on AI back-ends, currency unstated, against about 70% margins (Diego Roshardt · Diego Roshardt (Jul 2025) ▶ 10:42). For text products that still leaves room. Blake Anderson puts AI costs under 3% of revenue unless you do image generation or advanced voice (Blake Anderson · Blake Anderson (Dec 2024) ▶ 14:00), and GlowUp reports about 90% margins (Louis · Louis (Dec 2025) ▶ 10:44).

Generative media is the exception. Nicolai Klemke spends about $45,000 a month on GPUs and video APIs, “almost half” of revenue, with “not much profit in the end” (Nicolai Klemke · Nicolai Klemke (Jul 2025) ▶ 09:59).

Free usage is the classic trap. Bressler burned through about $5,000 of OpenAI costs in days after going viral (David Bressler · David Bressler (Feb 2024) ▶ 07:59).

(David Bressler · David Bressler (Feb 2024) ▶ 07:59)

Guests describe four responses:

• No free trial. Zafir offers none because top models aren’t cheap (CJ Zafir · CJ Zafir (May 2025) ▶ 12:39).
• Usage caps. Castmagic caps usage on a monthly plan and keeps checking model and transcription spend; the speaker is one of the co-founders (Ramon Berrios, Blaine or Justin) (unnamed guest · Ramon Berrios & Blaine (Jun 2024) ▶ 06:52).
• Bring your own keys. Praneeth Pike moved to BYOK in a desktop app so a one-time price could work, and reports about $80,000 in six months (Praneeth Pike · Praneeth Pike (Oct 2025) ▶ 04:03) (Zero-marginal-cost one-time-purchase AI app).
• Model routing. Flo uses OpenAI for easier tasks and Anthropic for deeper analysis (Flo · Flo (Apr 2026) ▶ 15:38).

Step 5 — Hedge platform and model dependence

Platform shocks are real. Tibo says TweetHunter was nearly killed by the X takeover while making 200K a month, which is why he now runs a portfolio (Tibo · Tibo (Dec 2025) ▶ 16:32). Saeed Ezzati says he has become “numb” to the risk to his ChatGPT extension; his fallback is to rebuild on other platforms (Saeed Ezzati · Saeed Ezzati (Aug 2024) ▶ 11:22). Lotts Ezeike’s WhatsApp bot hit about 15K a month in per-conversation fees before he moved it to his own apps (Lotts Ezeike · Lotts Ezeike (Nov 2025) ▶ 03:08).

To hedge the model itself, John Rush wires in three providers and lets users choose, because the best model changes monthly (John Rush · John Rush (May 2025) ▶ 12:37).

Where guests disagree

• Build on APIs you don’t control? Mike avoids it: AI can support the product but not be integral to it (Mike · Mike (Nov 2025) ▶ 10:58). Andy Cloke would rather use the models to power his business than build on them, because competition there is intense:

(Andy Cloke · Andy Cloke (Aug 2025) ▶ 06:45)

Yasser Elsaid concedes Chatbase was a wrapper and says he should have committed harder anyway (Yasser Elsaid · Yasser Elsaid (Jul 2025) ▶ 11:43).

• What is the moat? Brand and distribution for Nevo and Finn (Alex Finn · Alex Finn (Jun 2025) ▶ 11:55); workflow depth and features the model owner can’t copy for Berrios and Bressler (David Bressler · David Bressler (Feb 2024) ▶ 12:29).
• Will AI costs fall? One Prop GPT co-founder (Eyal or Yahli) says they keep falling (unnamed guest · Eyal & Yahli (Nov 2025) ▶ 09:06). Colin McIntosh says cheap credits are user acquisition and the window may close (Colin McIntosh · Colin McIntosh (Jan 2025) ▶ 02:46).

What can go wrong

• Revenue claims don’t hold together. Large consumer-AI figures are self-reported and sometimes inconsistent. Letterly’s “$250,000 a month” (Anton · Anton (Nov 2025) ▶ 01:30) is later given as “250,000 in revenue” with no period. The Sheets Resume transcript says $220,000 where the title says $20K a month (Colin McIntosh · Colin McIntosh (Jan 2025) ▶ 01:24). Stella’s “350, 360 MRR” with 12,000 payers probably means thousands (Sarah Pearl · Sarah Pearl (Jul 2026) ▶ 01:38). The Cal AI figures are given by one of the co-founders (Blake Anderson or Zach Yadegari) (unnamed guest · Blake Anderson & Zach Yadegari (Aug 2026) ▶ 02:27).
• Ethics are not discussed. AI “humanisers” are marketed to students for getting past detectors, and the episodes don’t discuss it (unnamed guest · Nikita & Evgeniy (Sep 2025) ▶ 02:08).

How strong is the evidence

The evidence badge is strong, but the topic dates quickly. Counting repeat guests (Blake Anderson, David Park, Nevo, Rob Hallam, Colin McIntosh) once and excluding host lines, the independent sources per step are roughly: magic moment or workflow (25), beyond the wrapper (10), distribution moat (9), pricing around cost (18), hedging (8).

• Step 3 is the most host-heavy, because Pat’s lines are excluded from the count.
• The platform-risk story rests mainly on Formula Bot (2024) and Tibo.
• Cost figures are one-off monthly self-reports, several with the currency unstated.

Go deeper

Consumer AI apps · AI SaaS for workflows · AI wrappers & moats · AI API costs & margins · AI agents & agent-ready SaaS · Vibe coding · Zero-marginal-cost one-time-purchase AI app · What’s your tech stack (and what does it cost)? · What’s the business model and how did you decide on pricing? · What are your costs and margins? · Services → productised → product