I am writing the post, because I want to show that with AI, really complex and meaningful projects can be done, if the proper tools were used the proper way.
I needed a project that I can use, so that I see how far I can go with only AI usage. I needed something doable, but relatively complex.
And then I remembered an idea my wife shared with me, a few years ago: an app that monitors her grandfather's behavior and sends a notification if he stops moving. at that time, AI integrations were not considered as an option, and static algorithms to model behavior (with adaptation) were complex and fragile. I am not that smart to implement such thing.
I started in December 2025. I read a lot, watched videos, I found a AI framework that I wanted to try: BMAD. The idea behind it felt known to me. Agile methodology, but with agents. What can go wrong!?
I had all I need. A language that I don't know, Android OS, that have changed a lot for the last 15 years , and AI models that I can use to dynamically adapt the behavior changes. A chosen framework. Looked complex enoungh for me. So many unknown things. The only thing I had was my experience in the software development!
I needed one week to go through the BMAD worklflow and get all documentations out of it. The main plan was to ask AI to implement the detection of the following concerns:
* Unusual stillness (potential fall or medical event)
* Did not wake up on time
* At an unfamiliar location at an unusual hour
* Phone silent for too long
I ended up with the following technical stack:
* Kotlin + Jetpack Compose + Material 3
* Room + SQLCipher for encrypted local storage
* Google Gemini for behavioral analysis
* Resend API for transactional email alerts
* Clean architecture: :domain (pure Kotlin) -> :data -> :ui -> :app
Having the basis known to me and the AI, the next five weeks the dev agents did the implementation.
And when I started testing, ALL FAILED. All was there, but nothing was connected. Like, 20 devs did something without a single daily meeting, ever!
I think that, if I tested more in the middle phases, knowing and checking the code regularly , investing more time in the initial BMAD workflow steps, I would have gotten better results. I also tried different models, before sticking permanently to Opus.
I didn't stop. Weeks of testing and fixing. What I learned the hard way: Android dev isn't as enjoyable as it used to be. Every OEM has its own way to kill my 24/7 monitoring
With AI and multiple LLMs, I ended up with an 11-layer service recovery system:
* Foreground service with IMPORTANCE_MIN channel
* WorkManager periodic watchdog
* AlarmManager backup chains
* BOOT_COMPLETED receiver
* SyncAdapter
* Batched accelerometer sensing
* Exact alarm permission recovery
* OEM-specific wakelock tag spoofing
* START_STICKY restart
* Safety net AlarmClock at 8-hour intervals
* User-facing gap detection with OEM-specific guidance
And new issues came up every day :D
And here are metrics:
422 prod files, 87k+ lines of prod code, 2251 tests, 53k+ line of testing code.
Was it worth it?!
With all the unknowns and mistakes, it took me 6 months. My investment so far is mostly the Claude Max subscription. Old-school, that's 2-3 people for the same 6 months.
Would I do it again? Yes. I know what to do now: pick a language I know, test more often, review the code often, write better docs. Use the right tools and LLM. Is the code quality good? Honestly, I don't care. As a PM, I care about fast delivery, low cost, good performance, stability, documentation. If the code were bad, I wouldn't get those. The app shipped and looks stable — so the code is in decent shape. And seeing how Claude Code works, the code is full of comments. Ordinary devs don't document this well. From that angle, the code is excellent!
The website (AI generated) : https://howareu.app/ Any advice, feedback, opinions will be highly appreciated.
