AI prototyping: how I actually build usable products with v0, Bolt, Replit & Lovable
A hands-on guide to AI prototyping that actually ships — from Figma to functional products in under 2 hours.
TL;DR? Here’s your first move:
Grab one of your Figma files
Write a 10-line PRD
Drop it into v0, Bolt, Lovable, or Replit
Make one small change at a time
Ship something real, fast
You don’t need perfect tools. You need a repeatable workflow.
I’ve tried just about every AI prototyping tool that popped up in my feed.
Some promised to turn prompts into products. Others swore they’d replace my entire dev team. Most of them left me with half-working buttons and broken layouts that couldn’t survive even a simple user flow.
But over the past 3 months, I found a workflow that actually works. One that helped me ship real, usable products with v0, Bolt, Replit, and Lovable.
If I had to teach someone where to start building with AI, this is exactly what I’d show them.
Start with context (and no, “start from scratch” isn’t a strategy)
AI doesn’t work in a vacuum. If you throw it a vague prompt like “make a dashboard,” it will hallucinate something out of 2007.
Here’s what works instead:
Best case: a Figma file.
If not: screenshots of an actual product.
Worst case: a wireframe sketch on paper or iPad. Anything is better than nothing.
Without visual context, AI makes blind guesses. You’ll waste more time correcting its creativity than building usable stuff.
Write a lightweight PRD (yes, even for AI)
This isn’t waterfall development. But AI still needs clarity.
My go-to is a product-spec.md file with key bullets:
Example PRD — MoodJournal (built with Replit):
# Core Features
- Daily journal entry with text, mood selection, and custom tags
- Weekly mood trends shown visually
- Local storage first, fallback to Supabase# UX Notes
- Minimal onboarding (2 screens max)
- Mobile-first UI with large tap targets
- Entry should take under 30 seconds
Result? You will be positively surprised… Play with it here or watch the video:
https://949db4e9-25d3-4caf-bf5e-929fb9fbb058-00-3u0hr8kdpq5te.kirk.replit.dev/
This single file helped Replit’s AI generate a working prototype that nailed every flow, from onboarding to mood tracking.
3. Prompt with precision (no “build me an app” energy allowed)
Talk to AI like a junior dev — specific and structured. Here’s a real-world example I used in Lovable:
Prompt — MealPrep Pro:
Create a weekly meal planner with a grid (7 days × 3 meals). Each card should show the recipe name, image, prep time, and tags (vegan, keto, etc). Tapping a card opens full details. Add a button to generate a shopping list based on selected meals.
Result? A fully interactive meal planning tool — built in minutes, ready to hand off.
https://preview--meal-grid-genius.lovable.app/
Iterate like a builder, not a perfectionist
Here’s the trap: you get a decent prototype, and start tweaking colors and spacing.
Don’t.
Instead:
Make one change at a time
Preview with real data
Roll it back if it doesn’t work
The perfect font doesn’t matter if the product doesn’t solve a real problem.
When I built “HabitStack” (a habit tracker), I ran 6 iterations in 45 minutes. From basic checkboxes → heatmaps → micro-animations → weekly summaries.
What I use — and when
v0 — Fast UI with clean code
Seamless Figma import
Real code export
Live preview, instant deploy
Bolt — For data-driven apps
Supabase integration
Terminal access
Great for CRUD-heavy tools
Replit — Full-stack prototyping playground
Built-in AI agent, console, collab mode
Ships to live URL
Lovable — Designer-friendly, no code needed
Visual UI, live preview
Figma support
Great for brand-safe asset generation
Stop chasing the perfect tool — master the workflow
Every AI demo looks magical. But tools don’t build great products — clarity does.
Once I locked in this workflow:
Context → PRD → Precise Prompting → Rapid Iteration
…everything clicked.
Now I ship prototypes in under 2 hours that used to take me 3 days. With higher quality.
Because I spend more time on real user needs — and less time fighting messy code.