A complete app in 34 days: 85 actual hours with AI versus roughly 800 estimated by hand
A backend developer built a complete app in 34 days and 85 actual hours with AI, versus roughly 800 estimated by hand: numbers and method from the git log.
Luca Di Domenico
· 3 min read
In short
A backend developer with eight years of experience, who had never used React Native, completed an entire application in 34 days working only in the evenings, on a $100-a-month subscription: 296 commits, roughly 58,000 lines of production code and 85 actual hours versus the roughly 800 estimated without AI. No vibe coding: architectural decisions made by the developer, specifications written before the implementation, tests dictated to the agent and line-by-line review of the generated code.
💸 A single $100-a-month subscription and a project built in the evenings, after work. For once, someone tracked every number instead of just shouting “10x”.
85 actual hours versus roughly 800 estimated by hand
The figure that completely changes the perspective is this: done manually, the same work would have taken roughly 800 hours. With AI, 85 were enough.
The story comes from a backend developer with eight years of experience who, before July 1st, had never used React Native.
In 34 days, working only between 6:00 pm and 2:30 am because of a day job, he completed an entire application.
The numbers from the git log
The numbers, straight from his git log:
- 34 days and 296 commits
- roughly 58,000 lines of production code
- 17 React Native screens, a NestJS server with Socket.IO and Redis, Solidity smart contracts, a Next.js site and Docker images built through CI
- 85 actual hours of his time
React Native or Flutter: a personal note
A small personal note: massive respect for the result, even though I hate React Native and happily remain team Flutter.
For those who don’t know them: they are both frameworks for building mobile apps. React Native uses the phone’s native components, while Flutter draws the entire interface itself.
But that’s a holy war we’ll leave for another post 😄
What it would have cost without AI
His estimate, without AI, is roughly 800 hours: the equivalent of 10-12 months of free evenings.
Design, moreover, is not his trade, and handing it to freelancers would have cost him between $15,000 and $30,000 (we’re talking US market).
“I didn’t vibe code”: the method
But the most important part is how he worked. His words:
“I didn’t vibe code.”
💡 Here’s what separates those 85 hours from a prototype destined to be rewritten:
1) He made the architectural decisions, not the agent.
2) He wrote the specifications before kicking off the implementation.
3) He told the agent exactly which tests to write.
4) He reviewed all the generated code, line by line.
The multiplier is the reasoning, not the subscription
The real multiplier comes from the reasoning you put in before and after the generation. The subscription is just the tool.
AI has made code almost free. The judgment needed to decide what to build and what to verify has stayed in your hands.
What changes when building software with AI
I see a lot of myself in what this developer shared.
Before, when building a software product, most of the time was absorbed by implementation, and less was left to reason about the decisions to make.
With AI this ratio has flipped: today we spend less time writing code and much more understanding what to build, how to build it and what to verify.
The last thing you shipped with AI: did you guide it this way, or did you let the model decide?
Want to know what this means for your product?
If you're building software with AI and want a second opinion on how you're working, tell me about your project.
Tell me about your project