Zero Code Experience: The Full Record of Building an iOS App in 90 Days with AI

Spent six months browsing job sites. Started nothing.

Bought three Python books. All unread. Purchased two video courses. Stopped at chapter three. For a long time, the assumption was that it was a motivation problem. Looking back, it wasn't. There was no visible path.

So the decision was made to build the thing that didn't exist. Navily—a learning app where answering a few questions generates a personalized roadmap and AI-written study materials.

This is the full record of a non-engineer, someone from the business side of things, building it solo with AI. Every number here comes from git history or invoices.

What Navily Is

Answer five questions. Get a goal-backward learning roadmap. Each chapter's study material is generated on the spot by AI—15 to 30 minutes per chapter.

This isn't a course recommendation engine. The material itself is AI-generated, tailored to the learner's context. That's the differentiation—though if a competitor replicates it, the moat is thin. That much is clear.

Why Build It

The desire was to change careers. But "what to learn" was opaque. Searching "Python self-study roadmap" returned 100 articles, each more confusing than the last. Six months passed without a single step forward.

The moment the path becomes visible, people move. They stay still because they can't see it.

At a previous job, a DX initiative was assigned. After a meeting full of terms like "API integration," a manager quietly asked: "What's an API?" Something was needed that could say "start here"—regardless of age or title. AI could fill that gap.

Tech Stack—A SaaS Running on ~$350/Month

Infrastructure cost with 29 users, all figures from actual invoices:

Supabase Pro (DB, auth): $25
Vercel Pro (hosting): $20
Claude API (content generation + chat): ~$100
Gemini API (image generation): free tier
Resend (email): free tier
Stripe (payments): 3.6% of revenue
Domain: ~$1/month

Frontend: Next.js 15, Tailwind, shadcn/ui. The selection criterion was singular: "Can one person operate this without breaking it?"

Three Failures and Three Wasted Weeks

Over 90 days, 30+ features were implemented. User feedback revealed that only three things actually mattered: the hearing-to-roadmap flow, AI study materials, and progress tracking.

Two weeks on a 3D bookshelf. One week on a knowledge graph. Three weeks poured into "nice to have."

"Nice to have" and "would miss if gone" are entirely different categories.

First login took two days. Connecting Supabase auth to Next.js produced redirect loops and vanishing sessions. The fix came from digging through Supabase's official example repo git history.

Study material generation took 8 minutes. SSE (Server-Sent Events) streaming brought it down to 5–7 seconds. Completely different user experience.

The 3D bookshelf was a judgment error. Built with Three.js—satisfying to build. User reaction: "So... where do I read the material?"

90 Days in Numbers

Commits: 1,000+
Tests: 1,097 (all passing)
Users: 29
Paying users: 7
Monthly cost: ~$350
Personal salary: $0
DB migrations: 162

Five Things a Non-Engineer Actually Needed

Redefining "technical skill"

It's not memorizing syntax. It's articulating what to build, communicating it precisely to AI, and verifying the output. Writing design documents and spotting when something is off mattered more than coding fluency.

Resting and returning

Development didn't happen every single day. The important part was coming back. The git commit log served as a streak counter—a small incentive not to let it go cold.

The decision not to build

AI builds whatever is asked. The ability to build becomes a trap. "Does the user actually need this?" has to be asked continuously. The answer is usually simpler than expected.

Using AI as a partner

AI handles information organization and code drafting. Design decisions and user understanding stay human. This division stabilized around month two.

Having something you want to build

This is the foundation. With insufficient technical skill but a clear "I want this thing to exist" drive, the gaps get filled—through research, through AI, through grinding through errors.

What's Next

Incorporation is planned for August 2026. The question to answer: "If Navily disappeared tomorrow, would you miss it?"

Nothing grand can be claimed yet. 29 users. $0 salary. But the look on someone's face during an interview when they said "this is what I've been looking for"—that feeling is probably not wrong. Or maybe it is. Moving forward regardless.

The full journey is documented on X. → @ai_in_reallife

To try Navily: → navilyai.com

If there is anyone out there—non-engineer, business background—who has something they want to build:

Ninety days ago, hearing "you're going to ship an iOS app" would have seemed absurd. It happened. Not because the ability was there, but because the desire to build was.

What is it that you want to build?

Tech Stack

Web: Next.js 15 / Tailwind CSS / shadcn/ui / Supabase / Vercel / Stripe
iOS: Swift 6 / SwiftUI
AI: Claude Sonnet (Anthropic) / Gemini (Google)
CI/Testing: 1,097 tests / GitHub Actions