paint-brush
Get Rich Building Verticalized AI Wrappers (Even If You Don’t Code)by@kamildebbagh
1,665 reads
1,665 reads

Get Rich Building Verticalized AI Wrappers (Even If You Don’t Code)

by Kamil DebbaghNovember 6th, 2024
Read on Terminal Reader
Read this story w/o Javascript

Too Long; Didn't Read

GenAI wrappers are making millions by packaging AI for specific use cases. You don’t need VCs or deep tech skills—just hustle, tools, and a focus on solving real problems. Start fast, monetize early, and consider defensibility later.
featured image - Get Rich Building Verticalized AI Wrappers (Even If You Don’t Code)
Kamil Debbagh HackerNoon profile picture

GenAI wrappers are quietly printing millions. Stop theorizing & start building—there’s still time to make a fortune.


When GPT-3.5 and ChatGPT dropped in December 2022, it opened floodgates for new opportunities in text, voice, image, video, and beyond.


Two tribes emerged: those who started building—the makers—and those who objected that building on top of AI models was not defensible—the thinkers.


Spoiler: tons of makers got rich.


It’s a new gold rush.


Opportunities are aplenty. Every corner of the web & the world is ripe for the taking.


And it’s not too late to join the wave. It’s even better than the 2010s SaaS era:

  • You don’t need VCs.
  • You (mostly) don’t need technical skills.
  • All you need is your hustle muscles and tools like GPTEngineer, Bolt, and Cursor.


Many of you are skeptics, and rightfully so, so let’s start with a list of successful GenAI wrappers from the GPT-era [1] ⬇️

Examples of success

Bootstrapped


  • Chatbase (web app) ~$4M ARR after 1yr (launched in February 2023) — “Build a custom GPT, embed it on your website and let it handle customer support, lead generation, engage with your users, and more.”
  • AragonAI (web app) ~$5M ARR in April 2024 (launched in Feb 2023) — “Get professional AI headshots in minutes with our new AI headshot generator”
  • PhotoAI (web app) ~$150K Monthly Revenue (founder started working on AI photos in late 2022) — “save money and use AI to do a photo shoot from your laptop or phone instead of hiring an expensive photographer”
  • RizzGPT (mobile app) ~$2.4M ARR — A seduction coach to help you write your messages on dating apps
  • Umax (mobile app) ~$5M ARR after 6mo (launched in Nov 2023) — a coach that rates your looks based on a few photos and offers personalized suggestions for improvement
  • Crayo.ai (web app) ~5M ARR after 8mo (launched in Nov 2023) — "Generate viral-ready clips in seconds”
  • Musicfy.lol (web app) ~1.5M ARR — “Sing Like Your Favorite Artist In Seconds With AI”
  • OurBabyAI (web app) 5-figures exit (launched in April 2023, acquired in June 2024) — “See Your Future Baby Instantly With AI Baby Generation”


VC-backed:

  • Harvey (web app) ~$25M ARR in July 2024 (launched in Jan 2022) — ChatGPT for lawyers


Other Hot Segments:

  • AI Voice Agents: Dozens of wrappers built on top of ElevenLabs or ChatGPT: they want to disrupt call centers, sales calls, insurance calls, etc.
  • Companionship apps: Think a chatbot that is your “AI best friend”, “AI girlfriend” or whatever. Loneliness is a real problem in 2024. These companies solve it. Character.ai did it first by building its own foundational model, it lets you talk to your favorite characters. And now a lot of companies are doing it by just being a wrapper on top of other models. They add a twist (e.g., being a mobile app, or living in a whatsapp conversation, or adding AI voice & AI images, etc.)


These examples show that success is possible—but how can you replicate it? It starts with understanding the common patterns among these winning products.

How did they do it?

These makers built something people wanted:

  • They verticalized AI models by packaging them for 1 specific use case (e.g., chatgpt for seduction)
  • Built a wrapper with a specific interface relevant to the problem (e.g., “upload an image of your dating app convo”),
  • And distributed the solution where their audience hangs out (e.g., subreddits of 18-25yo men).


Does it mean it’s easy and trivial? No. Building profitable software is still tough. But it’s never been easier.


Take RizzGPT. It seems trivial, but it works because it’s convenient. People won’t realize they can use ChatGPT for dating tips, but an app spoon-feeding the solution? Game on.


Vatsal Sanghvi’s sums it up:


Users don’t care whether you are building a wrapper or not, as long you solve a problem for them — Users will use your product


Of course, not everyone is convinced. As the success of these wrappers has grown, so have the criticisms. Let’s tackle the most common objections head-on.

Common Objections & Rebuttals

Objection: “There’s no moat”

Sure, GenAI tech alone isn’t defensible, but moats come later. Use the wow-effect as a lead magnet and monetize fast. Then expand your product with regular features to lock in users.


To see how this objection plays out in reality, consider these case studies of companies that adapted and thrived, despite the perceived challenges:


  • 2024 — Revid.ai by Tibo Louis-Lucas


    They generate videos with GenAI, building on top of a lot of common models / APIs.


    But they don’t stop here: they integrate with your social media accounts, so you can post directly from Revid. They also have a “worker” that automatically creates and posts videos for you every day. While these automations are not per se AI features, when combined with Revid’s GenAI, it creates a sort of agent, like a content creator who works for you and posts day and night. The combination is powerful. I suspect that these automation features were attempts to add layers of value on top of the GenAI, in order to increase the moat.


    Revid's "worker" feature


    These are just smart ways to extend the scope of the product in order to cover more of the user journey, i.e. what the user would do before/after the GenAI part. And these are great ways to reduce friction for your user, therefore increasing the value of your product.


    Start with a GenAI entry point, then build a workflow product to cover your user’s journey.


  • 2016 — Doctrine.fr (Google for lawyers, in France)


    I worked at Doctrine as a Product Manager and still am a shareholder — so I’m well-versed in the company history, and maybe biased 😁.


    Once upon a time there was a $10B market called “legal information” - databases of laws, court decisions, and other legal docs.


    In the 2010s, these expensive oligopolistic products looked like peak clunky 90s software.

    LexisNexis' infamous search form. Typical software for lawyers in the 2010s.


    In the age of Google, this status quo was growingly intolerable.


    Came the founders of Doctrine, who very astutely built a Google-like interface for legal information: a unique search bar, letting users search in natural language (👋 goodbye disgusting search forms).


    Doctrine's Doogle-inspired early landing page. Simple. Elegant. Efficient.


    Lawyers loved it. It felt like magic.


    Now here’s a secret: Doctrine was a simple wrapper on top of Elasticsearch, a search engine technology.


    Incumbents didn’t realize it — and clients didn’t care.


    This initial feature was enough for Doctrine to get a growing user base, revenue, and eventually funding.


    With the cash, Doctrine hired a bunch of brilliant engineers and built more tech-heavy features, developing its own in-house AI models.


    Doctrine started to also tackle more use cases, upstream or downstream of the user journey.


    Eventually, the company got acquired 8 years after its founding for 9 figures.


    In other terms, a simple wrapper allowed Doctrine to create a wow effect — and then they built a successful workflow product.



The playbook to build a successful wrapper is still the same, be it in 2016 with Elasticsearch, or in 2024 with GenAI models: in the early days, speed beats strategy.


But the “no moat” crowd isn’t entirely wrong either—AI wrappers face fierce competition, especially as more entrepreneurs spot the opportunity. The secret is understanding that you don’t need a moat to start, but you’ll need one to keep growing.


So you can stop after the first few millions and sell your business—but if you’re in for the long-run, to survive the next phase, you’ll need to think about defensibility. In B2C, this often means building a brand. In B2B, it’s about embedding into workflows, as explained by NFX.

Objection: “Who would use this? Anyone can just build their own RAG / use Langchain / use open source models from Hugging Face”

Reality check: most people aren’t techies, and even fewer will use open-source models from Hugging Face. Only 43% of 18-29-year-olds had used ChatGPT by Feb 2024 (source).


Source: https://www.pewresearch.org/short-reads/2024/03/26/americans-use-of-chatgpt-is-ticking-up-but-few-trust-its-election-information/


Yes, someone could copy-paste their Bumble chat into ChatGPT, but they won’t. Convenience wins, and techies underestimate how lazy or uncreative people are.

Objection: “GPT-5 will kill you”

Sure, it might. But as a wrapper, you’re adaptable. Swap models as better ones emerge—Claude 3.5 Sonnet (new), GPT-5, whatever. Let the AI giants play war games while you focus on serving users.


Companies that got crushed by ChatGPT (like Jasper) built basic wrappers on GPT-3. When ChatGPT (3.5) was released, they got steamrolled — to use Sam Altman’s lingo — because they weren’t adding value. Lesson: Make your product something that improves as foundational models improve, not a product that dreads the next OpenAI release.

Objection: “But Kamil most wrappers fail and building a successful one is very hard — you have a survivorship bias!”

Success is still hard, but building profitable software has never been easier:

  • You don’t need to convince a bank to lend you millions with your home as a collateral
  • You don’t need to pay ad agencies for expensive marketing campaigns
  • You don’t need to bend over backwards in front of VCs
  • You don’t need months to build physical stuff
  • You don’t need engineering skills
  • You don’t need design skills
  • You still need to find something people want
  • You still need to design and code something → AI will do 90% for you. The last 10%? You’ve got this. Google, GPT and GRIND.
  • You still need to distribute your product → organic content on social media and online communities, using AI & automations, are easier avenues to do it.
  • You still need to put in the hours, know your audience like no one does, be smart, copy, adapt, go further than others would.


But yes, failure is a likely option. Even the best indie hackers like Pieter Levels fail most of their ventures, with a 4/70 success rate over 11 years.


https://x.com/levelsio/status/1843731398009471015


Verticalized GenAI Wrappers are at the same time:

  • The new “dropshipping” for high-agency hustlers — a lot of these wrappers will be short-lived, more of hustles than real businesses. Typically for 16-25 yo builders who are eager to make their first 5, 6 or even 7 figures.
  • The starting point of amazing startup stories (like Harvey, or like 67% of the latest YC batch’s startups, S24, that are categorized as “AI”). Typically for more experienced builders who care about building a business for the next 10 or 100 years.


I loved this Twitter thread that explained why more and more Gen-Z founders prefer to bootstrap AI wrappers instead of looking for VC money:

https://x.com/deedydas/status/1842233563506422013

Objection: “All the wrapper ideas have already been done — now it’s only red oceans full of sharks”

I will give it to you, among the list of winners that I shared, a lot of them had a first-mover advantage. I wouldn’t recommend you to build a RizzGPT copycat today.


But you could be the next first-mover.


Because new technologies are released every other week.


Anthropic released its “Computer Use” API 10 days ago (Oct. 22, 2024), it uses computers the way people do—by looking at a screen, moving a cursor, clicking, and typing text.


OpenAI released a public real-time API for its advanced voice mode 30 days ago.


So many use cases that were impossible before are now solvable through a simple wrapper.


And so many more get unlocked every passing day.


Even without new technologies, there is an infinite number of industries, use cases, niches, that are still waiting for their GenAI savior app to come fix their problems.


Problems are everywhere. Hidden in plain sight. Waiting for you to solve them.


I was yet reminded of this underrated truth this week, while hanging out at a friend’s rooftop party in San Francisco.


I met with Alex Andrei, who’s co-founding Claritycare AI. Their founding story was quite simple: they were in the middle of an EntrepreneurFirst batch (startup incubator), looking for problems to solve in the healthcare space.


At the time, Andrei realized that his wife Emelia, a “back office pharmacist” for a health insurance company, was spending a lot of time manually reviewing customer’s requests asking the insurance to pay for their treatment.


Alex, being a technologist, had the idea to use an LLM to solve this problem. It worked like magic. This is how Alex & his co-founder decided to double down on this problem with his co-founder.


I loved this story because it shows that niche problems — with huge market sizes nonetheless :) — are everywhere.


Hidden in plain sight.


Waiting for the right person to discover and solve them.


This could be you. What are you waiting for?


I’m feeling generous, so I’m sharing lists of problems at the end of the article ⬇️

How to Do It Yourself (Even If You Don’t Code)

If I had to start an AI Wrapper from scratch, here’s how I’d do it:


  1. Find a Problem in a Paying Niche: Ideally one I know well. More on that below.

  2. Build It With AI Tools: You can literally build a full app (web/mobile) from scratch in a matter of hours with very little knowledge of code:

    1. Build with AI assistants:

      1. Full-stack builders (front, back, db, deployment): GPTEngineer, Bolt, Agent by Replit
      2. Front-end builder: V0 by Vercel
      3. Code assistants: Cursor’s Composer
      4. Check out this video by Greg Isenberg on how to use these tools
    2. Depending on the type of content I’m trying to generate, I’d use the APIs of:

  3. Create a Tailored Interface: I’d make it relevant to the user’s workflow.

  4. Monetize Day 1

  5. Distribute: I’d distribute the product on the relevant online communities (subreddits, discord servers, facebook groups) and, if B2C, heavily through Tiktok (UGC content creators, influencers deals & paid ads)

    1. Read this amazing article by Startup Spells on how the best GenAI wrapper B2C founders do their distribution (Arib Khan, co-founder of Musicfy.lol and Crayo.ai)
  6. Exit or Defend:

    1. If I want to play the short-term game, I would exit while revenue numbers look good
    2. If I feel like fighting against the inevitable copycats, I’d consider raising funds & investing in an engineering department to build workflow features and increase defensibility


While each of these points deserves a full article, I won’t pretend I have a magical recipe 😅. If you want to dive deeper into these topics, check out the further readings at the end of my article. I also strongly encourage you to read & watch content by the founders of the successful wrappers I shared. They’re very generous with their advice and often build in public.

List of problems

The best problem to pick is one you’re facing yourself — scratch your own itch. Otherwise, browse some of these resources for ideas:



Elon Musk dropped out in the 90s to launch a startup because the Internet wave was too big to ignore.


Today, GenAI is undeniable—will you ignore it?

THE END—go build now


About me

Hey guys, thanks for reading through my rants.


I’m Kamil. I’ve been building in AI for 5 years. I’m a 2x founder and currently Product Manager in a GenAI company: Poolday.ai lets mobile app marketers generate performing video ads in seconds.


I live in San Francisco & my favorite thing is to meet curious, ambitious, and contagiously enthusiastic people.


PS: my next article will be on how to mitigate non-deterministic outputs to build successful GenAI products — ping me if you’re interested about this topic!

Thanks to friends for reading drafts of this

  • Clovis Vinant-Tang, Yves Martin, Pierre-Geoffroy Pasturel, Valentin Foucault, Thibault Louis-Lucas, Antoine Dusséaux, Ghita Houir Alami, Paul Reffay, Paul Roussel, Mehdi Benbrahim.

Notes

[1] These figures are self-reported by founders. True or not, they’re directionally accurate.

Further Readings – Sources & Interesting stuff