You don't need a co-founder, a coding bootcamp or a five-figure budget to create an AI app anymore. One solo builder shipped a working AI fitness tracker for about $40 a month in subscriptions and roughly half a year of spare evenings — and by his own account, the code was never the hardest part. The part that actually slowed him down was everything around the code: the tool-hopping, the design, and the months he spent simply using his own app like a normal person at the gym.
What happened
It started with a spreadsheet that was too slow to use
The whole thing kicked off in March, when he first ran into "vibe coding" — a fancy way of saying you describe what you want in plain English and an AI writes the app for you. He'd been logging his workouts in Notion for about three months. It worked, but it was clumsy: mid-set, he wanted to punch in a weight and a rep count, not scroll through a table hunting for the right row.
There was one thing that setup did brilliantly, though. He could export the data — or just screenshot it and hand it to ChatGPT — and ask: how's my progress been lately, what's improving, what's stuck, what should I change in my program? He liked that far more than the built-in "AI coach" features in commercial fitness apps. So the idea was simple: build a real workout tracker for himself, but keep the ability to hand his own data to an outside AI.
The Lovable phase: magic first, mess second
He started with Lovable. The first impressions were almost magical — type what you want, and a while later something that resembles a real app appears on screen. Because he didn't yet know how to work with these tools, he'd cram 10 to 15 changes into a single prompt. Lovable would do some of them, break others, and he'd hit the free limit fast and wait for the reset.
The result was ugly. Some features didn't behave the way he asked. But it was a first MVP — minimum viable product, meaning the smallest version that's actually useful — and he could use it.
He didn't want to stay inside Lovable, though. The limits were small, and the database, the architecture and his ability to edit code with other AI tools were all locked to that one service.
Cursor and a real codebase
He moved the project to GitHub and switched to Cursor with the GPT extension. Development felt completely different: there was a proper codebase, he could see the diff (a side-by-side view of exactly what changed), roll back a bad change, and work across several files at once. Feature by feature, the app grew personal records, a rest timer, an exercise catalog, workout history and progress stats.
Functionally it kept getting better. Visually it was still rough.
$20 for Claude, and the design finally clicked
After a few YouTube reviews, he bought a Claude subscription for $20, and kept his $20 ChatGPT plan running too — $40 a month total. What interested him most was design.
Claude Design turned out to be surprisingly useful. Here's the odd part: plain Claude Code kept suggesting a mediocre interface, while Claude Design produced noticeably better results. The catch was the limits — a few prompts could burn through the entire five-hour allowance, far faster than regular Claude Code would. But for the first time, the app looked roughly the way he'd pictured it.
Then he walked away — on purpose
Active development stopped for a few months. He kept using the app himself, but he was busy with another project, and turning a personal tracker into a public product wasn't a priority.
That break may have been the most valuable part of the whole story. For months he used the app not as a developer hunting bugs, but as an ordinary guy at the gym. It got much clearer what he actually needed — and what he'd only added because it was possible to add it.
Back in late July: ship it
He returned to the project at the end of July. He saw yet another small habit tracker blowing up online and figured: if the app is nearly done and I use it myself, why not finish it properly and release it? It felt like only a little was left.
If you want to test ideas like this without paying for subscriptions up front, the free AI tools at MyKreaTool are a sensible place to start before you commit a single dollar.
What it means for you
You don't have to build a fitness app. The pattern here — describe it, generate it, move it into a real editor, polish the design, live with it for a while — works for almost anything you'd otherwise do in a spreadsheet.
At home
That household spreadsheet you've been maintaining — chores rota, recipe box, plant watering schedule — is a prime candidate. You describe the app you want in a prompt, and the AI builds a rough version you can start using the same day. No engineer required.
At work
Most teams have one ugly internal tool nobody wants to touch. If it's a small tracker or form, you can prototype a replacement in an afternoon and show it to your manager instead of describing it in a meeting. The lesson from this build: one change per prompt, not fifteen.
For a small business
A booking log, a simple client intake form, an inventory counter. The $40-a-month budget here is the whole point — you can build something narrowly useful for less than a single hour of a developer's time, then decide whether it deserves more.
For studying
A study tracker that exports cleanly to an external AI is genuinely powerful. You log your sessions, hand the data over, and ask what's slipping. That's the exact setup that started this whole project.
For creative work
Design is where paid tools earn their keep. Spending $20 on a subscription that produces a decent interface is often the difference between a prototype you hide and one you're happy to show people.
For extra income
Once an app is genuinely useful to you, publishing it — the builder here chose Google Play — turns a personal tool into something that could earn. Nothing about the stack he used required a team.
How to try it right now
1. Start free with Lovable. Go to the free tier, type one clear sentence about the app you want, and see what appears. Don't pay for anything yet.
2. Make one change per prompt. The single biggest mistake in this story was bundling 10 to 15 edits into one message. The AI did some, broke others, and it became impossible to tell what caused what.
3. Push it to GitHub. Once you have an MVP worth keeping, move the project out so it isn't locked inside one service. This is the step that gives you options later.
4. Open it in Cursor. Point Cursor at the repo and work with the GPT extension. You'll get diffs, rollbacks and multi-file edits — the things that make editing feel sane.
5. Add a design pass with Claude. If the app looks rough, spend the $20. Expect the five-hour limit to disappear quickly if you're iterating on visuals.
6. Use it yourself for a few weeks before you ship. Not as a developer. As a user. You'll cut half the features you were proud of, and that's the point.
Upsides and what changes
The obvious upside is money: $40 a month and evenings instead of a development budget. But the bigger shift is psychological. Once you've shipped one of these, the barrier between "I wish there was an app for this" and "there is now" collapses to a weekend.
You also get better at briefing. Working with AI tools is a communication skill, not a syntax skill — the people who get good results are the ones who can describe what they want precisely and check the output critically.
And there's a real strategic benefit to the pause. The builder's months of ordinary use taught him what his app needed far better than another sprint of feature-adding would have.
Limitations
Be honest about what this method doesn't solve. Free tiers run out fast, and every tool you touch has its own limits — Claude Design in particular can consume a whole five-hour allowance in a handful of prompts. The first output will look bad and behave worse, so your job isn't writing code, it's spotting what broke. Moving between tools costs real time: Lovable, then GitHub, then Cursor, then Claude is a lot of context switching for a solo project. And the "it's almost done" stage drags — this builder assumed only a little was left in late July, and that assumption is usually wrong. Finally, none of this replaces judgment about what to build. Half his features existed because they were easy to add, not because anyone wanted them.
Conclusion
The most useful takeaway from a half-year, $40-a-month build isn't the tool list. It's that the code was the easy part — the hard parts were deciding what to build, keeping prompts small, and using the thing long enough to know what deserved to stay.
Your action for today: open Lovable's free tier and describe the one spreadsheet you're most tired of maintaining, in a single sentence. That's it. Don't plan the revenue model. Just see what comes back.
Comments 0