Research project

The Vibe Coding Cliff

When you build exclusively with AI, what happens after the prototype?

prototype works some make it across
Take the survey 3 minutes · anonymous
Why this study exists

AI tools have opened software development to everyone.

Coffee shop owners, tradies, and parents on the couch can grab tools like Cursor, Lovable, Bolt, v0, or Replit and build solutions to problems they know best.

But the tools won't tell you when your product is ready for real customers, whether it can handle real load, or whether it'll hold up under security pressure. I'm researching where AI-built products hold up, where they fall apart, and what founders who clear the cliff do differently.

The hunch I'm testing

AI tools are great for getting a working prototype done quickly. But a working demo quietly skips the hard parts, whether your product will hold up in the real world, and whether the foundations are solid enough to build on. These cracks start to surface months after the prototype is working and can easily sink a startup. Some AI-built startups clear the cliff and I want to know what they're doing differently.

Who's behind it

A side project, run in my spare time.

I'm a software engineer. For the last 14 years my job has been taking software from "it works" to "it works reliably, securely, and at scale". This has often been in places where getting it wrong has real consequences, like banks and government.

If you've built with AI tools, your experience is what I'm after, doesn't matter if you sailed through or hit the wall hard. Founders who had no trouble are just as useful to hear from as the ones who got stuck.

David Craig

Principal Engineer, BNZ · Wellington

  • 14+ years across financial services and government
  • Web applications, platform stability, security & cloud architecture
View LinkedIn →

To be clear: This is a personal project run in my own time. It's not affiliated with or endorsed by my employer.

how to take part

Two ways to help.

01 / SURVEY

Fill out the survey

≈ 3 minutes · anonymous

Ten quick questions about how you built, what held up, and what didn't. Open to any NZ founder who's used AI tools to build their product.

Start the survey
02 / CONVERSATION

Have a proper chat

≈ 30–45 minutes · coffee or call

If you've been through it, I'd love to hear the full story. Relaxed, off-the-record by default chat, nothing gets published without your permission.

Connect on LinkedIn →
30 mins on me

Catch up for an interview and I'll spend half an hour reviewing your technical solution. See what's solid, what's shaky, what to watch for. No pitch, no strings.

findings first

Everyone who takes part sees the results before they're published. In time to spot your own cliff before you reach it.

$100 prezzy draw

Complete the survey and you're in the draw for a $100 Prezzy card.