I Have a Folder of Beautifully Architected Startups That Never Launched
Last weekend, I did something incredibly stupid.
I opened the "Projects" folder on my computer. Inside was a graveyard of good intentions. A social app with a perfect Kubernetes setup. An analytics tool with a microservices backend built for "massive scale." A niche marketplace with a database schema so elegant it could hang in a museum.
None of them ever got a single user.
I'm not alone in this particular brand of insanity. I asked some founder friends to confess their sins. The answers were an echo chamber of my own failures: "I spent a month on the CI/CD pipeline before writing a line of product code." "I re-platformed to a 'more scalable' framework twice before I even had a landing page."
We're all doing this wrong.
The Dirty Secret About "Building for Scale"
Here's the lie we tell ourselves: we're being responsible. We're "future-proofing."
Bullshit. We're procrastinating.
We're paying for a skyscraper foundation when we're only building a garden shed. All those debates about frameworks, databases, and container orchestration? It’s just theater. You don't need to handle a million concurrent users. You need to handle one user. Then ten. Then maybe a hundred.
The real kicker? I spent more time last year architecting than I did talking to customers.
That's not a strategy. That's engineering for an audience that doesn't exist.
What Actually Works (Based on Shipping, Not Speculating)
I decided to throw out my old playbook. No more building for a hypothetical future. No more choosing tech because it looks good on a resume. The only thing that matters is speed.
Here's what I learned.
1. Your "Default Stack" Should Be a Go-Kart, Not a Freight Train
Pick the stack that gets you from idea to a live URL the fastest. Not the one with the most theoretical power—the one that makes you think "I can ship this today" instead of "where the hell is the documentation for this?"
A clear winner has emerged for this:
Next.js + Supabase + Vercel. This isn't just a random collection of trendy tools. It's a system designed for pure, unadulterated speed.
Next.js is the frontend you don't have to argue with. It makes decisions for you so you can just build.
Supabase is your entire backend in a box. Database, auth, storage—it's all there in five minutes. Remember spending a week setting up user authentication? That's now a crime against productivity.
Vercel makes deployment so easy it feels like cheating. Push to Git, and it's live.
The test? If it takes you longer than 30 minutes to get from a blank folder to a deployed "Hello World," your stack is too slow.
2. Build Your "When-to-Break-the-Rules" Squad
Of course, one size doesn't fit all. But you don't need a new stack for every project. You just need a few "specialists" in your back pocket for when the job truly calls for it.
The "I'm a Wizard at This" Play: Got a decade of experience in Django or Rails? Use it! One founder I know still builds everything in Django and HTMX. Why? Because he can build a full-featured app in a weekend. Your personal fluency with a tool can often beat a "faster" tool you don't know.
The "One Codebase to Rule Them All" Play: If you absolutely need an iOS and Android app from day one, something like Expo (React Native) is your move. The business leverage of writing code once for web, iOS, and Android is insane for a solo founder. It's not a tech choice; it's a survival strategy.
The "Keep It Stupid Simple" Play: Can we talk about HTMX for a second? It’s a fascinating counter-punch to the complex JavaScript world. It's for people who realize their "web app" is basically just a series of forms and pages, and they don't need the complexity of a full-blown frontend framework. It’s a brilliant way to slash complexity.
3. The "AI" Reality Check
Let's get real about the robot in the room. Every developer's workflow now has an "AI feature" that's basically a souped-up autocomplete.
The argument I keep hearing is: "AI frees you from being tied to one stack!"
This is only half true. AI is incredible at writing the boilerplate you used to copy-paste. Need a login form with validation? Ask the AI. Need to connect to a new API? Ask the AI. It drastically lowers the cost of trying something new.
But it doesn't eliminate the need for expertise. It just changes it. An expert who knows their stack deeply can ask the AI better questions, spot the subtle bugs in the code it spits out, and actually architect the damn thing. "Garbage prompt, garbage code" is the new law of the land.
The lesson? Use AI as a bridge, not a crutch. Let it make you fearless in trying new things, but don't mistake its code for your own understanding.
My Latest Project (And The Time It Actually Took)
For my last idea—a simple B2B SaaS tool—here's how it broke down:
The Old Way (2019): Set up a server, install a database, write a REST API, configure a separate frontend app, build a CI/CD pipeline. Time to deployed MVP: ~2 weeks.
The New Way (Last Month):
npx create-next-app, connect to a Supabase project, push to Vercel. Time to deployed MVP: 25 minutes.
Result: I had a working prototype in front of a potential customer before my coffee got cold. All for the cost of a domain name.
The 48-Hour Test That Will Save Your Startup
Here's your homework:
Pick ONE default "Speed Stack" and commit to it for the next 3 months (the Next.js/Supabase/Vercel combo is a damn good start).
Take your most promising idea out of the "Projects" graveyard.
Give yourself exactly 48 hours to build and deploy the absolute most minimal version of it using your Speed Stack. No extra features. No "what if" architecture.
Ship it. Send the URL to at least one person.
Stop tweaking anything else until you get feedback from a real human.
The Truth About Building a Stack That Actually Ships
The best tech stack isn't the one that will scale to a billion users. It's the one that you actually use to launch something.
Your stack should feel like a shortcut to your customers, not a set of engineering puzzles you have to solve first.
The real question isn't "Which tech is better?"
It's "Which tech gets my idea out of my head and into the world before I lose interest?"
Start there. Your "Projects" folder will thank you.