What Happens When a Non-Coder Tries to Build an App with AI?

Cindy Adem
May 6
5 min read

Updated: 2 days ago

Spoiler: It worked— mostly. Here’s what I learned building Grace Notes Daily using only Replit’s AI tools.

I used Replit to build out Grace Notes. It is one of the most powerful tools for building apps by talking to AI

Introduction

When I set out to build Grace Notes, a Christian spiritual companion app, I had no idea what kind of journey I was beginning. I’m a product leader by trade; comfortable with shaping ideas, user experiences, and business strategy- but apart from some Python data analysis, I don’t know how to code. Not a single line.

What followed was an intense 8-day sprint using Replit’s AI assistant. This wasn’t just a passion project; it was a personal challenge:

Can someone with zero coding experience build a functional, AI-powered app with just plain English?

Watch this:

This is the story of how I did it, what I learned, and what I wish I had known before starting.

What is Grace Notes?

Grace Notes is a spiritual growth app designed to help users deepen their connection with God using modern technology.

Core Features Include:

AI-generated daily devotionals
Personalized “grace notes” with biblical references
A conversational spiritual companion
A journaling feature called “Heart Notes”
Prayer tracking and answered prayer celebrations

It’s intentionally calming and uplifting, with a green and gold color palette and an interface meant to be intuitive for all users; whether you're exploring the Christian faith for the first time or are spiritually mature.

The Build: 8 Days with AI, Zero Code

Every piece of Grace Notes was created by me, guided entirely by Replit’s AI. I never wrote a single line of code. Instead, I described what I wanted, iterated on what the AI gave me, and learned how to steer the assistant to meet my vision.

Some features came together smoothly. Others required frustrating backtracking when things broke or misunderstood prompts derailed progress. But every part of the build taught me something.

Integrations: What Stayed, What Got Cut

✅ What I Kept

OpenAI Integration: The engine powering our devotionals, grace notes, and spiritual conversations.
React + Tailwind: A modern frontend stack that helped deliver a polished, responsive UI.
LocalStorage: Used for data persistence in the demo version—simple but effective.
React-Confetti: A celebratory touch for answered prayers.

❌ What I Removed

Supabase Authentication: Initially part of the plan, but too complex for a quick demo.
PostgreSQL Database: Replaced with LocalStorage to reduce setup overhead.
Email Integration with SendGrid: Deferred to focus on in-app features.
User Preferences System: Scaled back to basic defaults to simplify UX.

The Challenges I Faced

🧩 Authentication Gone Wrong

When trying to implement Supabase authentication, the AI assistant broke existing functionality. It introduced changes without considering how they affected the codebase —highlighting a limitation of AI-generated code: it often works in isolation but struggles with interdependencies.

I had to spend several sessions just walking the AI back from errors it introduced— learning to describe bugs, even without understanding the underlying code.

🎉 Confetti and Chaos

Adding a simple confetti effect to celebrate answered prayers led to memory leaks and state bugs. The 'sprinkles' sometimes kept showing after navigating away or just didn't show up at all.

The AI struggled to trace the root cause. Eventually, with clearer prompts and lots of trial and error, we got it working— but this small feature proved how even “fun extras” can become technical landmines.

My Prompting Mistakes (and Fixes)

In hindsight, some of the biggest issues were caused by how I prompted the AI. Here’s where I went wrong—and what I learned:

Being Too Vague: Early on, I made requests like "implement authentication" without specifying enough details about how it should integrate with existing features.
Changing Direction Mid-Stream: I sometimes pivoted to new features before fully completing others, which made it harder for the AI to maintain context.
Not Providing Enough Context: When issues arose, I sometimes focused on symptoms rather than helping the AI understand the broader context of the problem.
Lacking Technical Vocabulary: Not being a developer, I sometimes struggled to describe technical issues in terms that the AI could easily understand, leading to misinterpretations.

A better approach would have been:

1. Clearer Feature Specifications: Outlining exactly what I wanted each feature to do and how it should interact with other parts of the app.

2. Iterative Testing: Testing smaller changes before moving on to the next feature.

3. Better Problem Descriptions: When issues arose, providing more comprehensive information about what was happening and what had changed.

4. Leveraging more Technical Terms: Some tech speak helps AI understand what we need better and behave accordingly.

Trade-Offs: From Full App to Demo

By day 6, I realized that trying to launch a full-featured product was slowing progress and increasing bugs. So I pivoted to a demo-first strategy; cutting anything non-essential and focusing on the core experience.

This shift:

Eliminated login complexity, allowing users to jump in immediately
Simplified data storage, avoiding database setup and cost
Accelerated delivery, helping me hit an 8-day MVP milestone

The end result? A functional demo that gets the vision across- without the tech debt.

Advice for Non-Coders Building with Replit

If you’re thinking about building an app with Replit and don’t know how to code, here’s what I’d recommend:

Start Simple: Begin with a minimal viable product and add complexity gradually. It's easier for the AI to help with incremental changes than to build complex systems all at once.
Learn the Basics: Even minimal understanding (what “state” or “component” means) can dramatically improve your prompting.
Be Specific: Vague requests confuse the AI. Detailed specs help it build what you’re imagining.
Test Constantly: Don't wait until a feature is "complete" to test it. Regular testing helps catch issues early when they're easier to fix.
Keep Notes: Write down what each feature does, how it works, and what bugs arise. This helps with context in future sessions.
Use Visuals: Share mockups or screenshots with the AI (Replit allows this!) to better convey UI expectations.
Understand the Limitations: The AI is powerful but not perfect. It's particularly challenging for it to understand complex existing codebases or make significant
Be Patient and Persistent: Building software is complex, and there will be setbacks. The key is to learn from them and keep moving forward.

Final Thoughts

At the time of writing, Grace Notes is about 8/10 complete in the dev environment. The core features are functional, the experience feels meaningful, and the content generated is genuinely impactful.

Deploying to Vercel has been harder than expected due to how the AI structured the project, but we're nearly there. I’ll update this post once the app is live.

Still, I’m proud of what this project represents:

Proof that non-coders can build powerful tools with AI
A spiritual product made with intention and creativity
A real-world test of what AI-assisted software development can—and can’t—do

Building with Replit has been a whirlwind—sometimes frustrating, often rewarding, always educational.

If you’re curious about creating your own app using only AI and determination, I hope this story encourages you to take the leap. If you use my link attached to the button below, I get a few extra credits to keep building, and sharing my journey with you.

Create your own Replit

Want to follow the journey or collaborate? Feel free to connect with me or drop a comment below. Screenshots and the live link coming soon. 🙏