⚡ I help (vibe) Coders build real Software 🎨 UI 🧠 backend , auth 💳 billing
Design × Code
Joined December 2024
- Tweets849
- Following88
- Followers164
- Likes352
Firebase vs Supabase — how I actually choose.
🟢 Supabase
• real Postgres
• RLS + SQL when the model is clear
• better once users, plans and invoices are tables
🔥 Firebase
• Auth in an afternoon
• Hosting is stupid cheap
• Extensions do the boring jobs
Stripe can sit on Firebase without safety risks
• Google AI on the same infra
• pay-as-you-go: you pay when people show up
• security rules + Google backbone
good enough for a solo SaaS
The split:
Product still moving? Firebase.
Data model already real? Supabase.
I don’t start with a VPS
so I can feel like a backend engineer (which I could be, because I used to code my own backends)
I start where hosting, payments, auth and AI comes together
don’t need three extra accounts.
Which team are you? Firebase or supabase
I am pro firebase
Most vibe-coded apps look random.
Not because the model is bad.
Because nobody planned the UI.
They open Claude and say “make it pretty.”
Then wonder why every screen feels like a different product.
I use Stitch for this now.
Most people still don’t.
How I plan UI before I touch code 👇
🧠 The part people skip:
Iterate inside Stitch.
Lock tokens early.
Export DESIGN.md.
PNGs are moodboards.
DESIGN.md is the actual UI.
If the radius and type change every prompt,
your app has no brand. Just vibes.
I stopped paying for Claude Code and Codex.
Not because they’re bad.
Because the bill got stupid.
Antigravity + Gemini 3.8 is the stack I actually finish SaaS with.
Why:
💸 pricing doesn’t punish long agent loops
🧠 3.8 is good enough for coding a whole application (frontend + backend) in one session
🧩 I can keep prompting instead of watching a 5-hour quota die
⚡ Flash speed > waiting on a “smarter” model that costs 5x
Claude still writes prettier diffs.
Codex still feels tight in the terminal.
I don’t care.
I ship more when I’m not scared to run the agent again.
If you’re vibe-coding a real product, optimize for volume of iterations.
Not the leaderboard.
Your AI just dumped 7 buttons into the bottom nav.
That’s how apps feel cheap.
Bottom nav is not storage.
It’s the skeleton of the product.
4 rules I use before I ship 👇
3/ Active state must be obvious in 0.2s
One change is not enough.
Do at least two:
• outline → filled
• color shift
• bolder label
If users ask “where am I?”, the nav failed.
7️⃣ The Contrast Effect 🏷️ Your brain evaluates every piece of information relative to the thing it saw immediately before. A 1,900 laptop in the cart? It feels like a rounding error. Never show costs in isolation!.