How much should you build before testing with customers?
Last Monday night, I had a hunch. By Wednesday afternoon, I had a working mobile app — and a lot more clarity.
I didn't build this to launch a company. I built it to test an idea, push myself technically, and see just how far I could get using AI tools I'd been experimenting with. Along the way, I was reminded of one of my core product beliefs:
You don't need a finished product to start learning. You just need something real enough for customers to react to.
Here's how I went from "I wonder if I could…" to a shareable demo in four days — and how this kind of scrappy experimentation is core to how I support teams through Align Product Studio.
The idea: recycling rules, made simple (with AI)
Earlier this year, I completed NYC's Trash Academy, a deep dive into the city's waste systems. What stuck with me most wasn't the infrastructure — it was the human behavior. Most New Yorkers are awful at sorting waste correctly.
DSNY (the NYC sanitation department) has a website that explains how to dispose of almost anything. But it's not exactly intuitive and you have to click through multiple pages to find the information you need. So I had a thought:
What if I could build a mobile app where you snap a photo and it immediately tells you how to appropriately dispose of that item — using image detection and NYC's own rules?
It felt like the perfect bite-sized test:
- Can I build a mobile app from scratch? (I never had)
- Can I learn to use advanced GenAI tools like Claude Code and integrate with Google's Vision API?
- Can I make the experience faster or easier than the tools available to NYC residents today?
I gave myself four days — with a soft deadline of Wednesday night, when my MVP Club meets for weekly check-ins.
The build: Claude, Expo, and a lot of curiosity
I've built several web tools so far, but never touched mobile. So I started from zero, using Claude Code to generate and refactor React Native code, and Expo Go to test everything in dev mode on my phone.
Tech stack + tools
- Frontend: React Native + Expo Go for sandbox testing
- Image Recognition: Google Vision API
- AI Support: Claude + Perplexity for planning, coding and debugging
- Database (later iteration): NYC recycling rules embedded locally
By Wednesday night, I had something functional. You could snap a photo, and the app would return instructions on how to dispose of it properly.
But it wasn't robust and it was often inaccurate. The recycling logic was too brittle as I realized I was building the rules independently. So on Thursday morning, I took a step back and asked Claude:
"What if we built NYC's rules into the app directly, rather than attempting to make assumptions after receiving Google Vision's results?"
That shift — combining image recognition with local logic — made the experience far more consistent and much more powerful.
You can see it here: Watch the 2-minute demo
The real goal: fast, tangible validation
I didn't care if this was "launch-ready." I cared about learning:
- Can I build it?
- Can it actually work?
- Does it unlock conversation or confusion?
- Could people use it as-is?
The answer to all four: yes-ish. And that's the point. Even a rough prototype gives you more signal than a Notion document ever will.
What it reinforced
Working > perfect
Pixel-perfect is irrelevant if the value prop isn't clear. This app wasn't sleek — but it was real.
AI is a legit accelerator
Claude wrote the code. Not always correctly, but enough to bring ideas to life quickly. Knowing how to prompt, debug, and refactor was the real superpower.
Builds spark better conversations than briefs
The moment I shared it, friends started tossing out use cases, edge cases, and new ideas. That's what you want early on.
Speed is a filter
Giving myself four days forced me to scope ruthlessly, prioritize function over finesse, and let go of anything that didn't directly support learning.
Why this matters
This wasn't just a fun side project. It's exactly the kind of work I do through Align Product Studio.
Teams can often get stuck trying to perfect ideas before they test them. I help them do the opposite:
- Scope a build that's just functional enough
- Move quickly with the help of AI tools and low-code "scaffolding"
- Validate assumptions with real user input — fast
Whether it's a prototype, internal demo, or pilot-ready MVP, I work with founders to go from fuzzy idea to working software that teaches them something useful.
If you're sitting on an idea — try this
- Give yourself a week
- Strip it down to the core interaction
- Build something real
- Share it with people who can give you feedback
Then see where it leads.
And if you want an experienced partner who can help you do it faster and with a strong customer-centric product POV — you know where to find me.