How I Build
The Stack.
A non-coder's toolkit for building real products. Honest notes on what works, what surprised me, and what I'd do differently. Not a tutorial — a practitioner's log.
The Tools
Six tools. One workflow.
The Workflow
How it all connects.
Honest Lessons
What actually surprised me.
Prompting is a skill, not a shortcut
The quality of AI output is entirely a function of how well you describe the problem. "Make a button" and "Make a primary call-to-action button in sage green that scales on hover with a subtle shadow" produce wildly different results. Investment in learning to prompt = compound returns.
Components before pages
Build the navigation and footer first. These are shared. Everything else is unique per page. I wasted time on landing pages with copy-pasted nav because I didn't know about components. Never again.
Your database schema is your architecture
Supabase tables are simple — but the decisions you make about structure shape everything downstream. Think about what data you're collecting and how you'll query it before creating the table.
Ship → observe → fix → ship
The cycle time matters more than the perfect build. A live site with three bugs teaches you more than a perfect local version nobody sees. GitHub → Vercel makes this loop fast enough to be useful.
Design system first, pages second
The most valuable thing I built was a consistent set of CSS variables (colours, fonts, radius, shadow) before touching any page. Changing the sage green now means changing one line, not 40.
Domain expertise is the actual differentiator
An AI tool doesn't know what a real estate sales team needs. I do. The intersection of "knows the domain deeply" + "can now build using AI" is where the competitive moat lives.
What I've Built
Experiments from the playground.
Curious about the process?
I write about building here — in the context of a non-coder founder navigating the AI engineering world.