I'm a DevOps/Cloud engineer, 6+ years in the field. My world is Python, Bash, Linux, CI/CD pipelines, cloud infrastructure — not mobile apps. Last October I started building a parenting app (ParAI) with a small team — though in practice, I handle most of the development myself. No funding, no big team.
7 months later: the app is live on iOS and Android, has 90+ blog posts, supports 11 languages, runs a NestJS backend with AI features on top. I have two kids (3.5 and 1.3) so my coding window is basically 9pm to midnight after they're asleep.
I had zero React Native experience when I started. Without AI coding tools this wouldn't exist — not just because of time, but because I was learning a completely new stack while building a production app.
The stack
- Mobile: React Native (Expo), TypeScript
- Backend: NestJS, MySQL, TypeORM
- AI features: Google Vertex AI
- Infra: GCP (DevOps background pays off here)
- AI coding: Kiro CLI with Claude, GitHub Copilot
How I actually use Kiro CLI
I'll be specific because most "AI workflow" posts are vague.
I use it as a thinking partner. When I'm about to build something, I describe it and let Kiro poke holes. "What about edge case X?" "Have you considered Y?" At 11pm my brain misses stuff. This catches it.
Boilerplate generation. Every new endpoint looks 80% the same. Controller, service, DTO, migration. I'm not typing that out for the 50th time. AI generates it, I fill in the actual logic.
Debugging when I'm too tired to think. Some nights I stare at an error for 10 minutes before realizing I should just paste it into Kiro. Usually gets me to the fix in under a minute. Embarrassing how often the answer is obvious once someone (something?) points it out.
Blog content. 90 posts didn't write themselves. I outline the key points, AI drafts, I edit heavily. The medical content (AAP guidelines, sleep schedules) I fact-check line by line. But the structure and first draft? AI handles that.
What I refuse to let AI do
Architecture. I've seen what happens when you let AI design your system — you get over-engineered nonsense with 14 abstractions for a CRUD app.
Auth and payments. No. I write these, I review these, I test these manually.
Infrastructure. 6 years of DevOps means I just do this faster myself. Explaining my GCP setup to an AI takes longer than just writing the config.
Real example from recently
Built the child module (ages 2-7) — the app originally only covered babies. The workflow:
- Described the scope to Kiro → it helped me break down what's different for toddlers vs babies (meal tracking instead of bottle feeds, behavior logging, screen time limits). Caught stuff I hadn't thought about — like how milestone tracking needs completely different CDC data for 3-year-olds vs 6-month-olds.
- Had it generate the new service layers, database migrations, and base UI screens
- I spent a few days wiring it into the existing system, making sure baby module users aren't affected, handling the switching logic between modules
- Asked it to write tests → got maybe 80% of what I needed, wrote the rest myself
- Reviewed, fixed edge cases, deployed
The whole module took about 2 weeks of evening sessions. Without AI? Probably 6-8 weeks at my pace.
The productivity difference (real numbers from my experience)
| What | Without AI | With AI |
|---|---|---|
| New API endpoint | ~2 hours | ~30 min |
| New screen | ~3 hours | ~1 hour |
| Blog post | ~2 hours | ~30 min |
| Translating to a new language | ~1 hour | ~15 min |
| Figuring out a weird bug | ~45 min | ~15 min |
Roughly 3-4x faster across the board. Some tasks more, some less.
The stuff nobody mentions
You build faster but you also build wrong things faster. I shipped 3 features in the first month that nobody used. Speed without talking to users is just wasted time.
AI-generated code has a "smell." After a while you can spot it — overly defensive, too many comments, weird variable names. I've started being more specific about style in my prompts.
It breaks your debugging muscle. I noticed after a few months that I was reaching for AI before even trying to understand the error myself. Had to consciously pull back. You still need to be able to debug without help.
Context window limits are real. Big features that touch 15 files? AI loses track. You need to break things into smaller chunks or it starts hallucinating about code that doesn't exist in your project.
Would I go back?
No chance. It's like asking if I'd go back to writing code without autocomplete. There's no going back.
But I want to be clear — AI didn't make me a better engineer. It made me a faster one. The decisions, the architecture, the product thinking — that's still 100% human. If you don't have those skills, AI just helps you produce garbage faster.
If you're doing something similar — solo dev, AI-assisted, shipping real products — I'm curious what tools you're using and what's working. Always looking to improve the workflow.
ParAI — free AI parenting app, ages 0-7. iOS and Android.
Top comments (0)