App building with Agentic AI
A look under the hood at how I’ve been using Flutter, Supabase, and Antigravity IDE to build a brand new product - and the challenges beyond engineering
Things have been a little quiet around here lately, but for good reason: I’ve been deep in the lab getting a new product off the ground.
Meet Klovva.
What is Klovva?
Without giving too much away before we officially launch, Klovva is a new mobile platform built to help modern parents share the mental load and tame the daily chaos of juggling the kids' schedules.
We’re still polishing things up behind closed doors (Hint: It doesn't look like the AI-generated image above!), but it’s been taking up the vast majority of my engineering hours outside of my contracting work.
Where AI Was Really Super Helpful
I know I'm not the first to say this, but AI is at its most useful when it acts as an accelerator for an engineer rather than a total replacement. Building Klovva has proved that point ten times over.
1. Product Designer > Syntax Writer
I didn't actually know Flutter before starting this project. In a traditional workflow, picking up a new cross-platform framework from scratch would have added weeks of googling, refactoring, and syntax bugs before getting a functional prototype running.
Instead, I acted more like a Product Designer and System Architect. I defined the high-level architecture, user flows, and product logic, while letting agentic AI handle the underlying UI code and widget state management.
2. Antigravity IDE + Gemini Pro
I built the bulk of the application inside Google’s Antigravity IDE, powered by Gemini Pro. Working with an agentic IDE felt like pairing with a hyper-fast junior developer.
However, despite what I wrote in a previous post, getting the new version (2.1) of Antigravity IDE to play nicely with my usual DevContainers setup did not go as smoothly as I would have liked.
Antigravity IDE 2.1 + DevContainers
- Push Everything into the Dockerfile: Don't rely on
devcontainer.jsonlifecycle scripts. Antigravity doesn't run post-scripts (likepostCreateCommandorpostStartCommand), so bake all CLI tools, Flutter SDKs, and system dependencies directly into your Dockerfile. - The "Container Started" Hang: If the launch hangs right when you hit the
"Container Started"status message, don't wait around. Disconnect from the container and immediately reconnect—it usually picks up the active session instantly. ::
3. MCP Integration with GitHub Issues
To keep the AI agent structured, I leveraged the Model Context Protocol (MCP) directly integrated with GitHub Projects and Issues. Instead of pasting long prompts into a chat window, I created structured issues in GitHub. The agent reads the issue context via MCP, generates an implementation plan, executes the changes across the repo, and closes the issue when done.
The Tech Stack
I selected a modern, serverless stack for getting the app off the ground quickly:
- Front-end: Flutter (App), Nuxt + Vue + TailwindCSS (Website)
- Back-end: Supabase (Postgres DB, Edge Functions, Auth)
- Hosting: Netlify
- Email Integration: Resend
- AI Integration: Gemini, Jina
- Analytics: PostHog
- Project Management: GitHub Projects + MCP
The Hard Part: From Code to Market
As engineers, we often fall into the trap of thinking that building the app is the main event. In reality, the engineering is the easy part (for us.)
Now, I’m turning my attention to things I don’t fully understand yet: Go-To-Market strategies, business incorporation, and monetization models. So whilst I have a vision for the app and where I want to take it, figuring out how to launch and scale a product is a whole new challenge - wish me luck!
Join the Waitlist
Back to Klovva - we aren't ready for a full release quite yet, but if you want to be among the first to test it out when we open the doors, go ahead and join the waitlist: