Back to the Drawing Board

So… I’m starting from scratch.

A year ago, I took my first crack at building this AI-based onboarding platform. I used Google Gemini as my co-pilot, organizing the project “Jira-style” with specific tracking codes to keep it honest.

It categorized tasks and recommended a tech stack:

Python Backend (My comfort zone).

TypeScript/Next.js Frontend (It pushed hard for this; as a backend specialist, I accepted it.)

We settled on a generic, “good enough” product vision and a crisp technical roadmap. It even generated visual mock-ups. I felt productive.

So, how far did I get?

I built the whole thing using generated code. It worked, but it was dreadfully ugly.

More importantly, it was brittle. The AI acted like a junior developer on a sugar rush—solving immediate problems while ignoring broader architecture.

The Terraform scripts left security groups wide open.

The Frontend was a spaghetti mess of components I was terrified to touch.

The “Boring” Vision killed my motivation the moment the code got hard.

Why I’m starting again

I realized that AI makes a terrible Architect.

It’s an incredible intern for boilerplate, but when it drives the vision, you end up with a generic, insecure toy.

What I’m doing differently

This time, I am the Architect.

Vision First: It’s not just “an onboarding form,” it’s a secure, “walled garden” tool. That comes from experience, not a prompt.

Component-First Frontend: I’m using Storybook + Cursor.ai to build isolated, testable components. This is how a backend mind survives in a frontend world.

Secure by Default: I design the VPC and Identity patterns myself; AI just fills in the syntax.

The goal is no longer just to “finish.” It’s to build something that survives the real world.

Back to the code editor.

— Brad

Disclaimer: This research is conducted in my personal time. All thoughts and opinions expressed here are solely my own and do not represent any current or former employers.