How We Built This (With AI): Building SnapFile with AI and a Two-Person Team

April 10, 2025
|
AI

When we started out building SnapFile, we were two full-time mid-career adults with some moonlighting help—and a whole lot of AI. Thanks to tools like ChatGPT, Cursor, Ideogram, and more, we were able to go from scratch to a launched product way faster than we expected. And while the business ultimately shut down, one of the biggest takeaways was just how much you can build today with a small self-motivated team—if you learn to use AI well.

Here’s what worked, what didn’t, and how we’d do it differently now.

The Magic of a Small Team + AI Leverage

I focused on the business side—setting up the business, writing copy, running marketing experiments, analyzing performance, being webmaster, etc. My co-founder handled the engineering. But AI seriously multiplied what both of us could do:

  • I used custom GPTs (ChatGPT) to 
    • Write website copy, blog posts, ad headlines, and help debug marketing campaigns
    • Answer detailed questions about the regulation and details of the law
    • Ask detailed questions about API documentation and other technical specs
  • I’d drop screenshots of our landing pages into ChatGPT to get quick, specific feedback on where we might be losing people and what to do about it.
  • I used AI to analyze Google Ads performance, dig through search term reports, and look for patterns in attribution.
  • I also used it to test out tons of headline variations and quickly get direction on messaging for different intents.
  • My co-founder, though not coding all day in recent years, was able to build and launch our entire product solo, leaning heavily on Cursor, Vercel, and GitHub Copilot. From backend logic to front-end layout to database setup—AI filled in the gaps.

AI didn’t just speed us up, it made a lot of this work possible in the first place.

So many GPTs

Unexpected Upside: Getting Hands-On with Technical Work

Before SnapFile, I wouldn’t have considered myself technical. But I learned to use a lot of AI, and it gave me just enough support and structure to take on things I’d normally avoid:

  • Used AI to help write and troubleshoot JavaScript on our website.
  • Had ChatGPT walk me through Google Tag Manager, Google Analytics, and Microsoft Clarity when I inevitably got confused. I took screen shots, asked a million questions, and always got a patient answer from the robot.
  • Parsed XML specs from a messy government PDF into something readable using ChatGPT and spreadsheets.
  • Helped manually QA our API integration by checking our code against the XML standards ChatGPT helped me organize.

Without AI, there’s no way I’d have tackled that kind of work. But thanks to robot help, I was able to jump in and actually be useful.

Thanks for the code, robot!

Where AI Fell Short (Then) — and What’s Changed

Even though we leaned heavily on AI, there were a few places where it just didn’t help much—at least not yet.

1. Design

We used DALL-E, and ChatGPT to brainstorm options, but the output wasn’t usable. It was messy and generic, and definitely not something we could use as-is. We ended up switching to Ideogram, Canva, and Figma—and eventually leaned on our moonlighting designer to make things look good. It wasn’t fast.

Bad AI generated images back then

Today? Tools like Vercel v0 and improved image generators are a lot more capable. You can get a clean layout and working components from just a few prompts—way more useful than when we started.

2. Website Building

We built our website in Webflow. It worked, but making changes was slow. Every update to layout or branding meant hunting through the Webflow UI and reworking sections manually. AI helped with the content, but the structure was painfully manual.

Today? You can use tools like V0 to prompt quickly into a complete, styled site—no template juggling required.

3. API Integrations

We had to integrate with a government API using strict XML standards—shared in PDF format. ChatGPT helped make the PDF more readable and helped us structure the spec, but we still had to hand-code everything. There was no way around the manual validation and debugging.

Today or Soon? We’re getting closer to LLMs that can reason through API schemas and even generate validated code for integrations. Back in 2024, AI couldn’t quite handle it, and it was a solid week of manual development work. 

Final Thoughts: A Different Kind of Team

SnapFile taught us that you don’t need a big team—you just need a capable one, supported by the right tools. It’s never been a more possible time to bootstrap your way into a software business.

If I were doing this again, I’d build pretty much in the same way: one senior engineer who’s great with AI tools, someone on the business side (like me) who can handle product, finance, ops, and a bit of marketing, and maybe a full time marketer. That’s enough to get a very good start.

More than anything, SnapFile was a crash course in building with AI. We had to build and launch an entire business with just two people, a stack of quickly changing AI tools, and no rules. It was an amazing forcing function. We had to get fluent with AI tools super fast—in product, engineering, marketing, and a pile of new technology will likely be how software gets built. It was an amazing experience to build a business, but the skills we built are arguably more important. We will be even faster next time!