There is a case study of a former optician who taught himself programming and is now earning about 3.5 million yen ($23,000) per month by building a personal SaaS using the AI coding tool "Claude Code."
When we hear "earned with AI," we tend to focus on "how it was made." But following his story, the most helpful part is "how it was sold." Now that building itself has become cheaper thanks to AI, the difference lies in "what to build" and "how to sell it." In this article, I will specifically introduce those two aspects with sources.
Distributing Selected Claude Code Skills
I've made it available at the end of this article.
Who is he and how much is he earning?
The protagonist is Samuel Rondot (X: @samuelrdt). Living in France, he was originally an optician. About 8 years ago, he taught himself programming through a 15-hour YouTube course and has since built several tools (his own post).
The revenue for one of the apps he created, "StoryShort" (a service that automatically generates short videos from text), is as follows:
- Last 30 days: $23,469
- MRR (Monthly Recurring Revenue): $22,309
- Total: $513,973
- Paid subscribers: 391
- Started: August 2024
(Source: trustmrr.com/startup/storyshort / as of July 5, 2026)
These figures can be verified because a service called TrustMRR connects directly to the Stripe (payment system) API to read revenue. This is actual payment data, not self-reporting. At an exchange rate of 150 yen to the dollar, it's about 3.5 million yen per month (rates fluctuate, so please check the current rate).
He also mentions that his "3 SaaS products total $28k–$35k/month," but this total is his own claim and figures vary by media. The $23k for StoryShort is what can be verified by third-party payment data. It's accurate to keep these separate.
The real lesson is the "Selling Method"
The most effective part of his method isn't the code or the AI, but the sequence of "identifying what sells first, then releasing it in a sellable form." This can be applied even if you can't write code. Let's look at it in order.
1. Verify "evidence of sales" before building
His rule is "only build things that already exist and show signs of success." Instead of coming up with ideas from scratch, he chooses genres that are already selling. His criteria for choosing a subject are (Indie Hackers interview):
- Is there search volume for that genre? (Are people searching for solutions?)
- Are competitors attracting customers via SEO or ads? (Using a tool like Ahrefs to check if competitor traffic is from search or ads)
- Is it something I would want to use myself?
- Is maintenance simple? (Can it be operated by one person?)
- Is the "path to money" clearly visible? If not, don't build it.
The important thing is to do this "before building." Many people build first and then get stuck thinking about how to sell, but he does the opposite, narrowing down to subjects with a basis for selling before moving.
2. Determine positioning by "improving just 1%"
Differentiation doesn't mean a grand invention. It's just crushing one frustration with existing tools. "Too expensive," "missing that one feature," "difficult UI," "no support for my language." Fixing any of these becomes a reason to switch. He consistently talks about the idea of choosing something existing that sells and making it slightly better (Indie Hackers). It's easy to understand his StoryShort as a slightly more user-friendly remake of the existing "AI short video" genre.
3. Don't "build then sell"; "test if it sells, then scale"
The sequence is as follows (from his own blueprint):
- Build a Minimum Viable Product (MVP) that works, even if it's rough.
- Immediately run Google and Meta ads to see the response (registrations or payments).
- Once you know people will pay, put in the effort to scale it.
The key is to use ads for "demand testing" rather than "post-completion customer acquisition." Use a small amount of ad spend to verify if people will pay for it; if there's no response, pull out quickly; if it hits, invest there. This sequence prevents spending months on development only to find "nobody wanted it."
4. Scale traffic with "SEO + Automated Content + Affiliates"
Once a hit is identified, shift customer acquisition to methods that are cheap and effective in the long run. In his blueprint, he writes about building up SEO and automated content, and further expanding through affiliates (a system where others introduce the product) (his own post). The flow is to hit with ads, then bring it to a state where it "runs even if ad spend decreases" through SEO and affiliates.
Key points for the building process
Once the selling method is decided, it's time to build. Although building has become cheaper with AI, knowing the "process" greatly changes how well you can make the AI work.
He says he lets "Claude Code do 90% of the work," but it's not a total hand-off. He also says, "Understanding code is important" and "I still learn code. AI just makes the first version faster" (Indie Hackers). However, there are tips for the "understand and operate" part that even beginners can grasp. Based on the stories of Celso Pinto (who built a SaaS alone with 713 commits, source) and Boris Cherny, the Anthropic engineer who created Claude Code (source), I've narrowed it down to four effective points.
Point 1: Don't let it build immediately; first write "what to build" in plain language
Something as rough as "A tool where entering text creates a short video. The screen only has an input field and a generate button" is enough. This becomes the design document.
Point 2: Place those premises in a "CLAUDE.md" file
This is like an instruction manual for the AI. By fixing what the project is and how to build it in one file, you don't have to re-explain it every time.
Point 3: Don't ask to "build everything"; request one small feature at a time
Divide it into "input field first," then "save function." Celso divided the work into about 50 small tasks and had the AI proceed one by one.
Point 4: Run it every time you build and have the AI fix it (Verification Loop)
If something is wrong, tell it "this part doesn't work like this" and have it fix it. Cherny says, "Creating a loop where the AI can verify its own results through tests or commands increases quality by 2-3 times."
In short, the hurdle to "build" is much lower than before. That's why the battle has shifted to the first half: the "selling method."
Recommended steps to try this
Since the selling method is the star, the steps also start with "finding something that sells." You don't need to write code yet.
Step 1: Choose one tool that is already selling and write down one thing "you would fix"
You don't need to come up with something from scratch. Anything you use daily and think "this part is inconvenient" is fine.
Step 2: Check if it has "evidence of sales" using the checks above
See if it's being searched for and if there are competitors. See if it's a genre where people are troubled and paying money.
Step 3: Leave the building part to the AI here
Install Claude Code (with the desktop app version, you can start without touching a black terminal screen) and try building a small version of that "one point you want to fix."
Step 4: Put out small ads or announcements and see the response
Don't aim for perfection immediately; first verify if "there are people who will pay money."
Among these, many people get stuck at Step 3, not knowing "what to type," so this time I have prepared a collection of Claude Code skills carefully selected by use case (ordered by what you should install first) and a collection of prompts you can copy and paste. Use them as tools to make the building part easier.
Conclusion
I want to tell you one thing calmly. To be honest, most people won't earn this much just by reading this. It's not a story of "anyone can earn if they do the same thing." Samuel is someone who has been working at it since 2017, and StoryShort is one product at the end of that accumulation. It costs money for ads, and it takes time to grow.
Still, the sequence of "identifying what sells first and leaving the building to AI" can be imitated starting today. Please start by writing down Step 1: "which tool's which part would you fix."
I am distributing tools for free on LINE to make the "building" part easier when you actually get to work.
- 50 Selected Claude Code Skills (by use case, starting with the essentials)
- 300 Prompts for Claude (starting with the top 10)
▼ Receive via LINE Registration
https://t.co/90omRA4UQ7
Please join the Open Chat, and then from the official LINE mentioned in the notes, send the word:
Skill
and you will be able to receive the benefits.
