It started, like many side projects do, with a moment of inspiration at a family event.
My nephew was turning one, and we wanted to livestream the party for family abroad. We ended up using a third-party app to stream to YouTube. It worked fine, but on the flight home, the developer side of my brain kicked in: That app was basically Encoder for YouTube Live. I could build that.
Then I had a second thought: Could I build it using AI, without actually writing any code myself?
I have been using AI / Agentic AI for sometime now. But using an Agentic AI to build a complete, end-to-end mobile app from scratch was a different kind of test. I didn’t want to build a weekend toy or a basic demo. I wanted to build a production app, publish it to the App Store, and see where the AI broke down.
The result is Vlogger Studio, a YouTube live streaming app that is currently live on the iOS App Store (with the Android version pending Play Store review). It’s a free YouTube streaming app with image stabilization and supports up to 4K resolution.
Here is what it actually looks like to build and ship an app with AI as developer.
The AI is a Developer
I made a rule for myself on day one: no manual coding. If I hit a roadblock, I couldn’t just open the file and fix it. I had to figure out how to explain the problem to the agent.
I know that the biggest trap with AI coding tools is treating them like magic boxes. If you type, “Build a YouTube streaming app,” you will end up with a messy, generic prototype that breaks as soon as you look at it.
Speed is easy. Production is hard. It’s not about how fast the AI spits out code; it’s about whether the pieces actually fit together when things go wrong.
Spec Driven Development
Before I wrote a single prompt, I spent time articulating the functional specification and design.
What should the app do?
- User Authentication options, linking YouTube account via OAuth, Checking Live enablement status, handling RTMP Streaming etc..
How should it work technically?
- Client-side OAuth with PKCE, Where to use YouTube v3 API, Firebase for user persistence and Cloud Functions for backend, Native iOS integration for camera and RTMP etc..
What are the constraints?
- Minimum iOS version, Native libraries for Camera and Streaming etc..
Built the plan first with spec and refined it as needed before handing it over to agent for development.
The Camera Bug (and Teaching the AI to “See”)
From that first prototype to a full app took about 1.5 weeks, spending a couple of hours a day. The AI was blazing fast at building out navigation, UI flows, and API wrappers and backend functions.
But then I hit the true bottleneck: physical hardware.
The app relies on the phone’s camera and microphone to encode and stream live video. You cannot test this in a laptop simulator; you have to build it and run it on a real iPhone.
At one point, the camera layout completely broke. The preview layout and orientation was wrong. This is where agentic AI struggles: Agent is not running in my iPhone. It couldn’t see what I was seeing.
To fix it without manually writing code, I had to build a feedback loop. I have asked agent to “Add structured logs to every single step of the camera initialization and view layout.”
I ran the app, reproduced the issue, copied the logs, and provided them back into Agent. The Agent analyzed the logs, formed a hypothesis about the view constraints, and generated a fix. We iterated through this exact loop—log, test, paste, and tweak— a few times until the camera perfectly adapted to orientation changes.
It was tedious, but it was a lightbulb moment. “Zero manual coding” doesn’t mean zero engineering. The work simply shifted from writing code to diagnosing systems and building feedback loops for the agent.
Step 4: The Other 50% of Shipping
After a week and a half, the app was technically done. I thought the finish line was in sight. I was wrong.
It took another full two weeks of work—none of which the AI could help much but accelerated a few — just to get the app ready for the public.
Shipping meant fighting through process: submitting videos to Google to get our OAuth scopes verified, proving our YouTube API usage, generating the exact right dimensions for App Store screenshots, writing privacy policies, and dealing with App Store review cycles. Relatively it was faster, it just took only 3-4 business days.
As iOS was ready, Android app was fast to build, but takes a mandatory 14+ days for production review on the Play Store for Individual accounts, which we are still waiting out, at the time of writing this post.
This is the part that gets left out of the “I built an app with AI in one hour!”. Generating code is fast. Shipping a compliant, secure, production-grade product is still a marathon.
What I Took Away
I started this project to see if you could build a production app without typing code. You absolutely can. But the experience totally reframed how I view the future of software development:
- Clear thinking is the new coding. The AI is brilliant at implementation, but it’s terrible at mind-reading. The more precise your requirements, the more useful the output. Spec-driven development is a superpower right now.
- You need to know the whole board. Because I had an end-to-end vision of how the mobile client, the backend, and the APIs needed to connect, I could guide the AI effectively. Without that architectural compass, it’s easy to let the AI paint you into a corner.
- The leverage is incredibly real. I spent exactly zero time agonizing over syntax errors, writing boilerplate UI components, or centering divs. I spent all my time on product logic, user flows, and testing.
Agentic AI doesn’t replace the need to understand how software works. If anything, it rewards developers and architects who can see the big picture. It changes us from being the ones laying the bricks, to being the ones designing the building and directing the crew.
And I have to say—watching Vlogger Studio app go live on the App Store without typing a single line of Swift or React Native? That was pretty cool.
I’d love to hear your experiences with AI. Feel free to connect or drop a comment!
Do not forget to try the app, and let me know what features you may want to see in the app.
https://vlogger-studio.harivemula.com/
Download on Android PlayStore (Coming Soon…)